MOPTA 2013 Modeling and Optimization:

Theory and Applications

August 14-16, 2013

Lehigh University, Bethlehem, PA, USA

1 !

Welcome to the 2013 MOPTA Conference!!

Mission Statement

The Modeling and Optimization: Theory and Applications (MOPTA) conference is an annual event aiming to bring together a diverse group of people from both discrete and continuous optimization, working on both theoretical and applied aspects. The format consists of invited talks from distinguished speakers and selected contributed talks, spread over three days.

The goal is to present a diverse set of exciting new developments from different optimization areas while at the same time providing a setting that will allow increased interaction among the participants. We aim to bring together researchers from both the theoretical and applied communities who do not usually have the chance to interact in the framework of a medium-scale event. MOPTA 2013 is hosted by the Department of Industrial and Systems Engineering at Lehigh University.

Organization Committee

Frank E. Curtis - Chair frank.e.curtis@lehigh.edu

Tamás Terlaky terlaky@lehigh.edu

Ted Ralphs ted@lehigh.edu

Katya Scheinberg katyas@lehigh.edu

Lawrence V. Snyder larry.snyder@lehigh.edu

Robert H. Storer Rhs2@lehigh.edu

Aurélie Thiele Aurelie.thiele@lehigh.edu

Luis Zuluaga

luis.zuluaga@lehigh.edu

Staff Kathy Rambo

We thank our sponsors!

!!!!! !!!!! !!! !!!! !!

2 !

Program

Wednesday, August 14 – Rauch Business Center 7:30-8:10 - Registration and continental breakfast - Perella Auditorium Lobby 8:10-8:20 - Welcome: Tamás Terlaky, Department Chair, Lehigh ISE - Perella Auditorium (RBC 184) 8:20-8:30 - Opening remarks: Patrick Farrell, Provost, Lehigh University - Perella Auditorium (RBC 184) 8:30-9:30 - Plenary talk - Perella Auditorium (RBC 184) Zhi-Quan (Tom) Luo, On the Linear Convergence of the Alternating Direction Method of Multipliers Chair: Tamás Terlaky 9:30-9:45 - Coffee break - Perella Auditorium Lobby 9:45-11:15 - Parallel technical sessions

Mathematical Models in Health Insurance Room: RBC 184

Chair: Aurelie Thiele

Computational Techniques for Smart Grids Room: RBC 271

Chair: Larry Snyder

Optimization Algorithms Room: RBC 91 Chair: Xi Bai

Robust Partial Capitation Aurelie Thiele

Security-Constrained Optimal Power Flow with Sparsity Control and Efficient Parallel Algorithms

Andy Sun

An Inexact Block-Decomposition CG Hybrid Method for Dense and Large-Scale Conic

Programming Camilo Ortiz

Robust Risk Adjustment in Health Insurance Tengjiao Xiao

Convex Quadratic Approximations of AC Power Flows

Hassan L. Hijazi

A Tight Iteration-Complexity Bound for IPM via Redundant Klee-Minty Cubes

Murat Mut Robust Value-Based Insurance Design

Shuyi Wang

Relaxations of Approximate Linear Programs for the Real Option Management of Commodity Storage

Selvaprabu Nadarajah

Risk Parity in Portfolio Selection: Models and Algorithms

Xi Bai 11:15-11:30 - Coffee break - Perella Auditorium Lobby 11:30-12:30 - Plenary talk - Perella Auditorium (RBC 184) Brian Denton, Optimization of Planning and Scheduling of Health Care Delivery Systems Chair: Aurélie Thiele 12:30-1:30 - Lunch - (RBC 292) 1:30-3:00 - Parallel technical sessions

Energy Management Systems Room: RBC 184

Chair: Miguel Anjos

Nonsmooth and Derivative-Free Optimization

Room: RBC 271 Chair: Frank E. Curtis

Optimization Under Uncertainty Room: RBC 91

Chair: Luis Zuluaga

MPC-Based Appliance Scheduling for Residential Building Energy Management Controller

Chen Chen

Full Stability in Nonlinear Optimization with Applications to Semidefinite Programming

Nghia Tran

Extensions of Scarf’s Max-Min Order Formula Luis F. Zuluaga

A Centralized Energy Management System for Isolated

Microgrids Daniel Olivares

Handling Equality Constraints in Expensive Black-Box Optimization Using Radial Basis Function

Surrogates Rommel Regis

Linear Solution Scheme for the Candinality Constarined Portfolio Allocation Models

Onur Babat

Optimization of Wind, Diesel and Battery Systems for Remote Areas Miguel Anjos

A BFGS-Based SQP Method for Constrained Nonsmooth, Nonconvex Optimization

Tim Mitchell

Computing Semiparametric Bounds on the Expected Payments of Insurance Instruments

via Column Generation Robert Howley

3:00-3:15 - Coffee break - Perella Auditorium Lobby 3:15-4:45 - Parallel technical sessions Large-Scale Optimization with Applications

to Machine Learning Room: RBC 184

Chair: Katya Scheinberg

Disruption Management Room: RBC 271 Chair: Lin He

Models for Electricity Market Mechanism Design

Room: RBC 91 Chair: Alberto Lamadrid

Sparse Inverse Covariance Matrix Estimation Using Quadratic Approximation

Cho-Jui Hsieh

Inventory Management for a Distribution System Subject to Supply Disruptions

Lin He

Environmental SuperOPF Electricity Market Planning Tool

Biao Mao A Deterministic Rescaled Perceptron Algorithm

Negar Soheili Azad

Optimal Dynamic Stochastic Scheduling with Partial Losses of Work Xiaoqiang Cai

A Nested Look-Ahead Model for Unit Commitment with Joint Ramping Capability

Requirements Boris Defourny

Complexity of Inexact Proximal Newton Method Xiaocheng Tang

A Modeling and Simulation Approach to Emergency Management Brian J. Hunt

An Experimental Study of Complex-Offer Auctions: Payment Cost Minimization versus

Offer Cost Minimization Rimvydas Baltaduonis

4:45-5:00 - Coffee break - Perella Auditorium Lobby 5:00-6:00 - Plenary talk - Perella Auditorium (RBC 184) Jorge Nocedal, Some Matrix Optimization Problems Arising in Machine Learning Chair: Katya Scheinberg 6:30-9:30 - Graduate student social - Graduate Student Center (Packer House)

3 !

Program

Thursday, August 15 – Rauch Business Center 8:30-9:00 - Continental breakfast - Perella Auditorium Lobby

9:00-10:45 - AIMMS/MOPTA Optimization Modeling Competition: Final presentations - Perella Auditorium (RBC 184) Chair: Peter Nieuwesteeg (winner will be announced at conference banquet) Team “OptNAR”, Universidad Politécnica de Madrid (TU-Madrid) and INTEC Raul Pulido Martinez, Natalia Ibañez Herrero (TU-Madrid), Adrian Marcelo Aguirre (INTEC); advised by Miguel Ortega Mier (TU-Madrid) Team “ORopt”, Technische Universität Berlin and ZIB Alexander Tesch (TU-Berlin); advised by Ralf Borndörfer (ZIB) Team “Universiteit Twente”, Universiteit Twente Corine Laan, Clara Stegehuis; advised by Bodo Manthey Team “ORTEC”, ORTEC, University of Amsterdam Harmen Boersma, Tristan Hands, Jan-Willem Arentshorst; advised by Frans van Helden 10:45-11:00 - Coffee break - Perella Auditorium Lobby 11:00-12:00 - Plenary talk - Perella Auditorium (RBC 184) Ignacio Grossmann, Relaxations for Convex Nonlinear Generalized Disjunctive Programs and their Application to Nonconvex Problems Chair: Ted Ralphs 12:00-1:00 - Lunch - (RBC 292) 1:00-2:30 - Parallel technical sessions

Approximation Algorithms Room: RBC 184

Chair: Amir Ali Ahmadi

Models for Uncertainty and Demand Response in Smart Grids

Room: RBC 271 Chair: Larry Snyder

Applications of MINLP Room: RBC 91

Chair: Hande Y. Benson

Rounding by Sampling Arash Asadpour

Optimizing Locations for Wave Energy Farms Under Uncertainty

Larry Snyder

Interior-Point Methods within a MINLP Framework Hande Y. Benson

Approximation Algorithms for Graph Partitioning Problems using SDP Hierarchies

Ali Kemal Sinop

Generation and Storage Dispatch in Electricity Networks with Generator Disruptions

M.Mohsen Moarefdoost

Mixed Integer Nonlinear Programming for Multi Vehicle Motion Planning: Ground and Underwater

Vehicles Pramod Abichandani

Approximation of the Joint Spectral Radius via Dynamic and Semidefinite Programming

Amir Ali Ahmadi

Modeling Demand Response for FERC Order 745 Yanchao Liu

Multiperiod Portfolio Optimization with Cone Constraints and Discrete Decisions

Umit Saglam 2:30-2:45 - Coffee break - Perella Auditorium Lobby 2:45-4:15 - Parallel technical sessions

Mixed Integer Optimization and Applications

Room: RBC 184 Chair: Ted Ralphs

PDE-Constrained Optimization Room: RBC 271

Chair: Jason Hicken

Facility Layout Problems Room: RBC 91

Chair: Abdul-Rahim Ahmad

Evasive Flow Capture: Optimal Location of Weigh-in-Motion Systems, Tollbooths, and Safety Checkpoints

Nikola Markovic

A Matrix-Free Augmented Lagrangian Algorithm for Large-Scale Structural Design

Andrew Lambe

A Novel Adaptive Boundary Search Algorithm for Solving Facility Layout Problems

Abdul-Rahim Ahmad Column Generation and Accelerating Schemes for

Mixed-Mode Aircraft Sequencing Problems Farbod Farhadi

Inexact and Truncated Parareal-in-time Krylov Subspace Methods for Parabolic Optimal Control

Problems Daniel B. Szyld

Cyclic Facility Layout Problem: A Hybrid Exact/Heuristic Optimization Approach

Abdullah Konak

Three Dimensional Knapsack Problem with Vertical Stability and Pre-Placed Boxes

Hanan Mostaghimi Ghomi and Walid Abdul-Kader

A Flexible Iterative Trust-Region Algorithm for Nonstationary Preconditioners

Jason Hicken

Solving the Unequal Area Facility Layout Problem: An Effective Hybrid Optimization Strategy Coupled

with the Location/Shape Representation Sadan Kulturel-Konak

4:15-4:30 - Coffee break - Perella Auditorium Lobby 4:30-5:30 - Plenary talk - Perella Auditorium (RBC 184) Omar Ghattas, The Stochastic Newton Method: Combining Large-Scale Optimization and Markov Chain Monte Carlo Methods for the Solution of PDE-Constrained Bayesian Inverse Problems Chair: Frank E. Curtis 6:00-7:00 - Cocktail reception - Asa Packer Dining Room (University Center) 7:00-9:30 - Conference banquet and competition results - Asa Packer Dining Room (University Center)

4 !

Program

Friday, August 16 – Rauch Business Center 8:30-9:15 - Continental breakfast - Perella Auditorium Lobby

8:45-10:45 - Parallel technical sessions Optimization and Differential Equations

Room: RBC 184 Chair: Yunfei Song

Healthcare Applications Room: RBC 271

Chair: Jackie Griffin

Advances in Portfolio Management and Pricing

Room: RBC 91 Chair: Elcin Cetinkaya

On the Complexity of Steepest Descent for Minimizing Convex Quadratics

Clovis Gonzaga

TBD Walter Massey

Full Characterization of Disjunctive-Conic-Cuts for Mixed Integer Second Order Cone Optimization

Julio Góez Preconditioners for PDE Constrained Optimization

Ekkehard W. Sachs Simulation Model for the Analyses and Cost Estimates of Combination HIV-Prevention

Strategies for the Elimination of HIV Chaitra Gopalappa

Implementing Real-Time Pricing in Wholesale Electricity Markets

Jingjie Xiao

A Primal-Dual Active-Set Algorithm for Large-Scale Convex Quadratic Optimization

Zheng Han

Patient-Bed Assignments in Hospital Systems Jackie Griffin

Portfolio Risk Management with Moment Matching Approach

Elcin Cetinkaya Convex Sets as Invariant Sets for Linear Systems

Yunfei Song Efficient Learning of Donor Retention Strategies

for the American Red Cross Bin Han

Robust Manager Allocation for Investment Management Yang Dong

10:45-11:00 - Coffee break - Perella Auditorium Lobby 11:00-12:00 - Plenary talk - Perella Auditorium (RBC 184) Henry Wolkowicz, Taking Advantage of Degeneracy in Cone Optimization with Applications to Sensor Network Localization and Molecular Conformation Chair: Luis Zuluaga 12:00-1:00 - Lunch - (RBC 292) 1:00-2:30 - Parallel technical sessions

Networks Room: RBC 184

Chair: Eric Landquist

Demand Systems Room: RBC 271

Chair: Miguel Anjos

Recent Advances in Sparse Linear Programming Room: RBC 91

Chair: Robert Vanderbei Multi-Agent Information Routing Under Dynamic and

Uncertain Conditions Dimitrios Papadimitriou

Scheduling of Multiproduct Pipelines for Transporting Liquid Fuels

Arun Sridharan

Estimating Sparse Precision Matrix by the Parametric Simplex Method

Haotian Pang Dynamic-Programming-Based Link Assignment for

Data Collection in Wireless Sensor Networks Yanhong Yang and Huan Yang

Consumer Demand Systems Based on Discrete-Continuous Models

Walter Gomez

Fast-Fourier Optimization Robert Vanderbei

A Simple and Efficient Strategy for Solving Large Generalized Cable-Trench Problems

Eric Landquist and Francis Vasko

Piecewise-Constant Regression with Implicit Filtering

Sanjay Yadav

Online PRSM Xingyuan Fang

2:30-2:45 - Coffee break - Perella Auditorium Lobby 2:45-4:15 - Parallel technical sessions Optimizing Supply-Demand Match in Power

Systems Room: RBC 184

Chair: Alberto Lamadrid

Semidefinite Optimization Room: RBC 271

Chair: Hongbo Dong

Optimization, Information, and Complexity

Room: RBC 91 Chair: Eugene Perevalov

Adaptive Load Management: Scheduling and Coordination of Demand Resources in Power Systems

Jhi-Young Joo

The Trust Region Subproblem with Non-Intersecting Linear Constraints

Boshi Yang

On the Connection Between the Reliability of Systems and the Notion of Invariance Entropy

Getachew K. Befekadu Co-Optimization of Grid-to-Vehicle Charging and

Ancillary Services Jonathan Donadee

Finding Hidden Cliques and Dense Subgraphs via Convex Optimization

Brendan Ames

Multiresolution Gaussian Process Model for the Analysis of Large Spatial Data Sets

Soutir Bandyopadhyay The Effects of Bulk Electricity Storage on the PJM

Market Roger Lueken

Conic Relaxations for Convex Quadratic Optimization with Indicator Variables

Hongbo Dong

On Optimal Information Extraction from Large-Scale Datasets

Eugene Perevalov

5 !

Program Highlights

8:30am-9:30am – Zhi-Quan (Tom) Luo, plenary talk (see page 9)

11:30am-12:30pm – Brian Denton, plenary talk (see page 6)

5:00pm-6:00pm – Jorge Nocedal, plenary talk (see page 10)

6:30pm-9:30pm – Graduate student social

9:00am-10:45am – AIMMS/MOPTA Optimization Modeling Competition: Final presentations (see page 12)

11:00am-12:00pm – Ignacio Grossmann, plenary talk (see page 8)

4:30pm-5:30pm – Omar Ghattas, plenary talk (see page 7)

6:00pm-7:00pm – Cocktail reception

7:00pm-9:30pm – Conference banquet and competition results

!

11:00am-12:00pm – Henry Wolkowicz, Plenary talk (see page 11)

Wednesday, August 14

Thursday, August 15

Friday, August 16

6 !

Speaker Biography

Dr. Brian Denton is an Associate Professor in the Department of Industrial and Operations Engineering at University of Michigan, in Ann Arbor, MI. Previously he has been an Associate Professor in the Department of Industrial & Systems Engineering at NC State University, a Senior Associate Consultant at Mayo Clinic in the College of Medicine, and a Senior Engineer at IBM. He is a Fellow at the Cecil Sheps Center for Health Services Research at University of North Carolina. His primary research interests are in optimization under uncertainty and applications to health care delivery and medical decision making. He completed his Ph.D. in Management Science at McMaster University, his M.Sc. in Physics at York University, and his B.Sc. in Chemistry and Physics at McMaster University in Hamilton, Ontario, Canada. Title: Optimization of Planning and Scheduling of Health Care Delivery Systems! Date: Wednesday, August 14, 11:30am-12:30pm Abstract: Optimization of planning and scheduling decisions under uncertainty is important in many service industries to increase the utilization of resources, match workload to available capacity, and smooth the flow of customers through the system. It is particularly important for healthcare delivery where applications include scheduling of patients to outpatient clinics, design of operating room schedules, and allocation of resources within healthcare facilities. In this talk I will discuss stochastic optimization models for scheduling services in outpatient procedure centers and hospitals. I will discuss three related problems. The first involves setting individual procedure start times for a single operating room (OR) given uncertainty in the duration of procedures. The objective of this problem is to minimize a weighted sum of three competing criteria: patient and OR team waiting time, OR idle time, and overtime. The second problem involves the allocation of surgeries across multiple ORs with the goal of balancing the fixed cost of opening ORs with the expected cost of total overtime. The third problem involves setting optimal arrival times for patients to an outpatient procedure center comprising multiple activities including: intake processes, surgery, and recovery. For each problem I will describe the model, stochastic optimization methods that can be applied, and numerical results based on real data to illustrate the potential impact of the model. I will also discuss open questions and future research opportunities related to optimization of health care delivery systems. !!!!!!!!!!!!!

Brian Denton

Associate Professor Industrial and Operations Engineering University of Michigan btdenton@umich.edu

7 !

Speaker Biography !

!

Dr. Omar Ghattas is the John A. and Katherine G. Jackson Chair in Computational Geosciences, Professor of Geological Sciences and Mechanical Engineering, and Director of the Center for Computational Geosciences in the Institute for Computational Engineering and Sciences (ICES) at The University of Texas at Austin. He also is a member of the faculty in the Computational Science, Engineering, and Mathematics (CSEM) interdisciplinary PhD program in ICES, serves as Director of the KAUST-UT Austin Academic Excellence Alliance, and holds courtesy appointments in Computer Science, Biomedical Engineering, the Institute for Geophysics, and the Texas Advanced Computing Center. He earned BS, MS, and PhD degrees from Duke University in 1984, 1986, and 1988. He has general research interests in simulation and modeling of complex geophysical, mechanical, and biological systems on supercomputers, with specific interest in inverse problems and associated uncertainty quantification for large-scale systems. His center's current research is aimed at large-scale forward and inverse modeling of whole-earth, plate-boundary-resolving mantle convection; global seismic wave propagation; dynamics of polar ice sheets and their land, atmosphere, and ocean interactions; and subsurface flows, as well as the underlying computational, mathematical, and statistical techniques for making tractable the solution and uncertainty quantification of such complex forward and inverse problems on parallel supercomputers. He received the 1998 Allen Newell Medal for Research Excellence, the 2004/2005 CMU College of Engineering Outstanding Research Prize, the SC2002 Best Technical Paper Award, the 2003 IEEE/ACM Gordon Bell Prize for Special Accomplishment in Supercomputing, the SC2006 HPC Analytics Challenge Award, and the 2008 TeraGrid Capability Computing Challenge award, and was a finalist for the 2008, 2010, and 2012 Bell Prizes. He has served on the editorial boards or as associate editor of 12 journals, has been co-organizer of 12 conferences and workshops and served on the scientific or program committees of 40 others, has delivered plenary lectures at 23 international conferences, and has been a member or chair of 20 national or international professional committees. Title: The Stochastic Newton Method: Combining Large-Scale Optimization and Markov Chain Monte Carlo Methods for the Solution of PDE-Constrained Bayesian Inverse Problems Date: Thursday, August 15, 4:30pm-5:30pm Abstract: We address the problem of quantifying uncertainties in the solution of ill-posed inverse problems governed by expensive forward models (e.g., PDEs) and characterized by high-dimensional parameter spaces (e.g., discretized heterogeneous parameter fields). The problem is formulated in the framework of Bayesian inference, leading to a solution in the form of a posterior probability density. To explore this posterior density, we propose several variants of a so-called Stochastic Newton Markov chain Monte Carlo (MCMC) method, which employs, as an MCMC proposal, a local Gaussian approximation whose covariance is the inverse of a local Hessian of the negative log posterior, made tractable via randomized low rank approximations and adjoint-based matrix-vector products. We apply this Stochastic Newton method to several large-scale geophysical inverse problems and study its performance. This is joint work with Tan Bui-Thanh, Carsten Burstedde, Tobin Isaac, James Martin, Noemi Petra, and Georg Stadler.

Omar Ghattas

John A. and Katherine G. Jackson Chair in Computational Geosciences Professor, Department of Geological Sciences, Jackson School of Geosciences Professor, Department of Mechanical Engineering University of Texas at Austin omar@ices.utexas.edu

8 !

Speaker Biography

! Prof. Ignacio E. Grossmann is the Rudolph R. and Florence Dean University Professor of Chemical Engineering, and former Department Head at Carnegie Mellon University. He obtained his B.S. degree in Chemical Engineering at the Universidad Iberoamericana , Mexico City, in 1974, and his M.S. and Ph.D. in Chemical Engineering at Imperial College in 1975 and 1977, respectively. After working as an R&D engineer at the Instituto Mexicano del Petróleo in 1978, he joined Carnegie Mellon in 1979. He was Director of the Synthesis Laboratory from the Engineering Design Research Center in 1988-93. He is director of the "Center for Advanced Process Decision-making" which comprises a total of 20 petroleum, chemical and engineering companies. Ignacio Grossmann is a member of the National Academy of Engineering , Mexican Academy of Engineering, and associate editor of AIChE Journal and member of editorial board of Computers and Chemical Enginering, Journal of Global Optimization, Optimization and Engineering, Latin American Applied Research, and Process Systems Engineering Series. He was Chair of the Computers and Systems Technology Division of AIChE , and co-chair of the 1989 Foundations of Computer-Aided Process Design Conference and 2003 Foundations of Computer-Aided Process Operations Conference. He is a member of the American Institute of Chemical Engineers, Sigma Xi, Institute for Operations Research and Management Science, and American Chemical Society. Title: Relaxations for Convex Nonlinear Generalized Disjunctive Programs and their Application to Nonconvex Problems Date: Thursday, August 15, 11:00am-12:00pm Abstract: This talk deals with the theory of reformulations and numerical solution of generalized disjunctive programming (GDP) problems, which are expressed in terms of Boolean and continuous variables, and involve algebraic constraints, disjunctions and propositional logic statements. We propose a framework to generate alternative MINLP formulations for convex nonlinear GDPs that lead to stronger relaxations by generalizing the seminal work by Egon Balas (1988) for linear disjunctive programs. We define for the case of convex nonlinear GDPs an operation equivalent to a basic step for linear disjunctive programs that takes a disjunctive set to another one with fewer conjuncts. We show that the strength of relaxations increases as the number of conjuncts decreases, leading to a hierarchy of relaxations. We prove that the tightest of these relaxations, allows in theory the solution of the convex GDP problem as an NLP problem. We present a guide for the generation of strong relaxations without incurring in an exponential increase of the size of the reformulated MINLP. We apply the proposed theory for generating strong relaxations to a dozen convex GDPs which are solved with a NLP-based branch and bound method. Compared to the reformulation based on the hull relaxation, the computational results show that with the proposed reformulations significant improvements can be obtained in the predicted lower bounds, which in turn translates into a smaller number of nodes for the branch and bound enumeration. We then briefly describe an algorithmic implementation to automatically convert a convex GDP into an MILP or MINLP using the concept of basic steps, and applying both big-M and hull relaxation formulations to the set of disjunctions.

Finally, we address the extension of the above ideas to the solution of nonconvex GDPs that involve bilinear, concave and linear fractional terms. In order to solve these nonconvex problems with a spatial branch and bound method, a convex GDP relaxation is obtained by using suitable under- and over-estimating functions of the nonconvex constraints. In order to predict tighter lower bounds to the global optimum we exploiting the hierarchy of relaxations for convex GDP problems. We illustrate the application of these ideas in the optimization of several process systems to demonstrate the computational savings that can be achieved with the tighter lower bounds.

Ignacio Grossmann !Rudolph R. and Florence Dean University Professor of Chemical Engineering Carnegie Mellon University grossmann@cmu.edu

9 !

Speaker Biography !

!!Zhi-Quan (Tom) Luo is a professor in the Department of Electrical and Computer Engineering at the University of Minnesota (Twin Cities) where he holds an endowed ADC Chair in digital technology. He received his B.Sc. degree in Applied Mathematics in 1984 from Peking University, China, and a Ph.D degree in Operations Research from MIT in 1989. From 1989 to 2003, Dr. Luo was with the Department of Electrical and Computer Engineering, McMaster University, Canada, where he later served as the department head and held a senior Canada Research Chair in Information Processing. His research interests lie in the union of optimization algorithms, data communication and signal processing. Dr. Luo is a fellow of IEEE and SIAM. He is a recipient of the IEEE Signal Processing Society's Best Paper Award in 2004, 2009 and 2011, as well as the EURASIP Best Paper Award and the ICC's Best Paper Award in 2011. He was awarded the Farkas Prize from the INFORMS Optimization Society in 2010. Dr. Luo has chaired of the IEEE Signal Processing Society's Technical Committee on Signal Processing for Communications and Networking (SPCOM) during 2010-2012. He has held editorial positions for several international journals, including currently being the editor-in-chief for IEEE Transactions on Signal Processing. Title: On the Linear Convergence of the Alternating Direction Method of Multipliers Date: Wednesday, August 14, 8:30am-9:30am Abstract: We analyze the convergence rate of the alternating direction method of multipliers (ADMM) for minimizing the sum of two or more nonsmooth convex separable functions subject to linear constraints. Previous analysis of the ADMM typically assumes that the objective function is the sum of only two convex functions defined on two separable blocks of variables even though the algorithm works well in numerical experiments for three or more blocks. Moreover, there has been no rate of convergence analysis for the ADMM without strong convexity. In this work, we establish the global linear convergence of the ADMM for minimizing the sum of any number of convex separable functions. This result settles a key question regarding the convergence of the ADMM when the number of blocks is more than two or if the strong convexity is absent. It also implies the linear convergence of the ADMM for several contemporary applications including LASSO, Group LASSO and Sparse Group LASSO without any strong convexity assumption. Our proof is based on estimating the distance from a dual feasible solution to the optimal dual solution set by the norm of a certain proximal residual.

!! !!!!!!

Zhi-Quan (Tom) Luo

Department of Electrical and Computer Engineering ADC Chair in Digital Technology University of Minnesota, Twin Cites luozq@ece.umn.edu

10 !

Speaker Biography !

! Jorge Nocedal is a professor in the Industrial Engineering Department at Northwestern University. His research interests are in optimization algorithms and their application in areas such as machine learning and energy management. His current research is being driven by a collaboration with Google Research. Jorge is passionate about undergraduate education; he was one of the developers of the ‘’Engineering First’’ Curriculum at Northwestern that exposes students to engineering design in their freshman year. He is currently the Editor in Chief for the SIAM Journal on Optimization, is a SIAM Fellow, and was awarded the 2012 George B. Dantzig Prize. Title: Some Matrix Optimization Problems Arising in Machine Learning! Date: Wednesday, August 14, 5:00pm-6:00pm !Abstract: The research presented in this talk is motivated by three applications: recommendation systems, speech recognition, and the training of vary large neural nets. In all these applications there is a need to solve large nonlinear optimization problems in which the unknown is a matrix. We describe state-of-the-art methods for solving these problems, and illustrate their performance using realistic data sets.

!!!!!!!!!!!!!

Jorge Nocedal Director of Optimization Center Professor Electrical Engineering and Computer Science Northwestern University nocedal@eecs.northwestern.edu

11 !

Speaker Biography !

!!Henry Wolkowicz is currently a professor in mathematics, in the department of combinatorics and optimization at the University of Waterloo in Canada. Prior, he was a professor at the University of Delaware and the University of Alberta. He received his Ph.D. from McGill University in Mathematics in 1978. Dr. Wolkowicz’s research deals with applications of optimization and matrix theory to algorithmic development for both continuous and discrete optimization problems. His research interests include: optimization in finite dimensional and abstract spaces; linear, nonlinear and semidefinite programming; matrix eigenvalue problems; and numerical analysis of algorithms. His combinatorial optimization work applies convex relaxations to hard combinatorial optimization problems. The relaxations are based on Lagrangian duality, and in many cases they result in Semidefinite Programming relaxations. Dr. Wolkowicz was chair for the SIAM Activity Group on Optimization (SIAG/OPT) from 2001-2004 and the SIAM Council from 2005-2011. He is the Associate Editor of the SIAM J. of Optimization; Math. Progr. B; J. of Computational Optimization and Applications , COAP; J. of Combinatorial Optimization, JOCO; Optimization and Engineering, OPTE; American J. of Mathematical and Management Sciences and has been organizer of several conferences and workshops. Dr. Wolkowicz has held several visiting research positions at Universite Paul Sabatier, Princeton University, Emory University and the University of Maryland. Title: Taking Advantage of Degeneracy in Cone Optimization with Applications to Sensor Network Localization and Molecular Conformation Date: Friday, August 16, 11:00am-12:00pm Abstract: The elegant theoretical results for strong duality and strict complementarity for linear programming, LP, lie behind the success of current algorithms. However, the theory and preprocessing techniques that are successful for LP can fail for cone programming over nonpolyhedral cones.

Surprisingly, many instances of semidefinite programming, SDP, problems that arise from relaxations of hard combinatorial problems are degenerate. (Slater's constraint qualification fails.) Rather than being a disadvantage, we show that this degeneracy can be exploited. In particular, several huge instances of SDP completion problems can be solved quickly and to extremely high accuracy. In particular, we illustrate this on the sensor network localization and Molecular conformation problems.

Henry Wolkowicz Professor Department of Combinatorics and Optimization The University of Waterloo hwolkowi@uwaterloo.ca

12 !

AIMMS/MOPTA Optimization Modeling Competition 2013

The fifth AIMMS/MOPTA Optimization Modeling Competition is a result of cooperation between Paragon Decision Technology (the developers of the AIMMS modeling system) and the organizers of the MOPTA conference. Teams of two or three graduate students participated and solved a problem of critical importance to hospital organizations. The teams were asked to consider an Operating Room (OR) manager’s task of scheduling and sequencing surgeries in a set of ORs, where, besides the inherent complexity of typical scheduling problems, OR scheduling is further complicated by the uncertainty of the time required to perform surgical procedures (including preparation, surgery, and clean-up times). The teams were asked to develop a tool to handle the scheduling and sequencing of surgeries in a hospital that aims to reduce the downtime for an OR, waiting time for a surgeon, and overtime for the OR staff, all of which create costs for the hospital organization. The teams had to form a mathematical model of the problem, implement it in AIMMS, solve it, create a graphical user interface, and write a 15 page report for the project. We are happy that 11 teams from 7 different countries participated in the competition. The panel of judges (Robert Storer and Luis F. Zuluaga from Lehigh University and Peter Nieuwesteeg from Paragon Decision Technology) selected the following three teams for the final:

Team “OptNAR”, Universidad Politécnica de Madrid and INTEC Raul Pulido Martinez, Natalia Ibañez Herrero (TU-Madrid), Adrian Marcelo Aguirre (INTEC)

advised by Miguel Ortega Mier (TU-Madrid)

Team “ORopt”, Technische Universität Berlin and ZIB Alexander Tesch (TU-Berlin)

advised by Ralf Borndörfer (ZIB)

Team “Universiteit Twente”, Universiteit Twente Corine Laan, Clara Stegehuis

advised by Bodo Manthey

Team “ORTEC”, ORTEC, University of Amsterdam Harmen Boersma, Tristan Hands, Jan-Willem Arentshorst

advised by Frans van Helden The finalist teams will each give 25 minute presentations (20 minute talks + 5 minutes for questions) on their work on Thursday starting at 9:00am in the Perella Auditorium. The winning team will be announced at the conference banquet on Thursday evening. One other team has received honorable mention for their work:

Team “PolytHEC”, École Polytechnique de Montréal and HEC Jean Bertrand Gauthier (HEC), Antoine Legrain, Étienne Beauchamp (École Polytechnique de Montréal)

Advised by Louis-Martin Rousseau (École Polytechnique de Montréal) We thank all the teams for their participation. We believe that it has been a very positive experience for all parties involved in the process.

Detailed Abstracts(Alphabetical by Speaker’s Surname)

Speaker : Pramod Abichandani (Drexel University, pva23@drexel.edu)Title : Mixed Integer Nonlinear Programming for Multi Vehicle Motion Planning: Ground and Underwater VehiclesAbstract : Mixed Integer Nonlinear Programming (MINLP) techniques are increasingly used to address challenging problems in

robotics, especially Multi-Vehicle Motion Planning (MVMP). In this talk, we present recent work in the area of Multi-Vehicle Path Coordination (MVPC) under communication connectivity constraints. We focus on two different groupsof vehicles — the first being ground based vehicles and the second group being underwater vehicles.For the ground based robots, we present technical approach and experimental results for MVPC. In this case, each ve-hicle robot starts from a fixed start point and moves toward a goal point along a fixed path, so as to avoid collisions andremain in communication connectivity with other robots. To the best of our knowledge this is the first experimentalimplementation of a real-time MINLP framework for solving MVMP.For the underwater robots, we present novel acoustic communication connectivity constraint formulations for theMVPC problem. These constraints account for the attenuation due to signal propagation and delays arising frommulti-path propagation in noisy communication environments, and specify inter-vehicle connectivity in terms of asignal-to-noise ratio (SNR) threshold. Simulation scenarios including up to 4 robots are simulated to demonstrate(i) the effect of communication connectivity requirements on robot velocity profiles, and (ii) the dependence of thesolution computation time on the communication connectivity requirement.In both scenarios, the optimization improved connectivity at no appreciable cost in journey time (as measured by thearrival time of the last-arriving robot). Results also demonstrate the responsive nature of robot trajectories to safetyrequirements with collision avoidance being achieved at all times despite overlapping and intersecting paths.

Speaker : Abdul-Rahim Ahmad (Rowe School of Business, Dalhousie University, Canada, Ahmad@dal.ca)Title : A Novel Adaptive Boundary Search Algorithm for Solving Facility Layout ProblemsAbstract : A survey of literature in the Facility Layout Planning area indicates that, despite the reported effectiveness of analyt-

ical algorithms, very few analytical methods have been published in the last decade. The paper focuses on the openspace facilities layout planning involving modules with constant aspect ratios. The paper presents a definition for“Local Optimum Layout” and introduces a “Near-Optimality Hypothesis”. Based on these definitions, a hybrid algo-rithm is presented that ensures convergence to a Local Optimal Layout at each iteration cycle. This algorithm is anovel combination of steepest descent and corner search. The algorithm is basically analytical with construction-cum-improvement hybridization and heuristics. The construction cycle places a new module at the optimal locationon the boundary of previously formed cluster of modules. The improvement cycle moves each module to its optimallocation in the direction of steepest descent and heuristically removes and resulting overlaps. The improvement cyclealternates boundary search and steepest descent moves until convergence to the Local Optimal Layout. Layout solu-tions produced by the proposed algorithm for well-known test problems were found superior to the published layoutsas well as the best layouts produced by commercial layout design software VIP-PLANOPT.

Coauthor(s) : Imran A. Tasadduq (iatasadduq@uqu.edu.sa) , M.H. Imam (mmimam@uqu.edu.sa)

Speaker : Amir Ali Ahmadi (IBM Watson Research Center, a_a_a@mit.edu)Title : Approximation of the Joint Spectral Radius via Dynamic and Semidefinite ProgrammingAbstract : The joint spectral radius (JSR) of a finite set of matrices is a nonnegative number that characterizes the maximum

growth rate that can be achieved by multiplying the matrices in arbitrary order. It is a natural generalization of thenotion of the spectral radius of a single matrix and has numerous applications across applied mathematics. Unlikethe spectral radius, however, the JSR is notoriously hard to compute; e.g., the decision problem “Is the JSR less thanor equal to one?” is algorithmically undecidable.In this talk, we give an introduction to the joint spectral radius and its applications. We then present algorithms basedon dynamic and semidefinite programming that either compute the JSR exactly for special cases or approximate itwith guaranteed accuracy in the general case.Based on joint works with Parrilo, and with Jungers, Parrilo, and Roozbehani.

13

Speaker : Brendan Ames (Institute for Mathematics and its Applications, bpames@ima.umn.edu)Title : Finding hidden cliques and dense subgraphs via convex optimizationAbstract : Identifying sets of densely connected nodes in graphs plays a significant role in a wide range of applications, such as

information retrieval, pattern recognition, computational biology, and image processing. We consider the problem ofidentifying the densest k-node subgraph in a given graph. Although the original combinatorial problem is NP-hard,we show that the densest k-subgraph can be recovered from the solution of a particular convex relaxation for certainprogram inputs. In particular, we establish exact recovery in the case that the input graph contains a single plantedclique plus noise in the form of corrupted adjacency relationships.

Speaker : Miguel Anjos (Polytechnique Montreal, anjos@stanfordalumni.org)Title : Optimization of Wind, Diesel and Battery Systems for Remote AreasAbstract : Imagine a mine in Nunavut (Canada) that is 1000 km away from the closest electrical grid. The high extension cost of

the grid means that electrical energy must be produced locally and autonomously. Traditionally electricity was sup-plied by diesel generation. This technology is easy to implement but is also expensive. The use of wind turbines inremote areas to reduce fuel consumption was proposed in the 1990s. This technology is now widely used in off-gridsites. The recent developments in storage technology mean that batteries may further reduce the use of diesel gen-erators. The challenge in optimizing such hybrid energy systems is to find both the optimal sizing and the optimaloperational strategy: the two are linked and impact each other. Because hybrid systems are often designed by simu-lation, dispatch rules must be set a priori, and this necessarily influences the outcome. We present an integer linearoptimization model to find the optimal design and dispatching scenario without the need for dispatch rules. The bestimplementable rules are then deduced from the optimal solution. Because our solution represents a perfect dispatch,it provides a reference to benchmark dispatch strategies.

Coauthor(s) : Thibault Barbier (thibault.barbier@polymtl.ca) , Gilles Savard (gilles.savard@polymtl.ca)

Speaker : Arash Asadpour (NYU Stern School of Business, aasadpou@stern.nyu.edu)Title : Rounding by SamplingAbstract : Linear Programming relaxation is a widely used approach to solve combinatorial optimization problems. The caveat

however, is to round the fractional optimal solution of the LP formulation to a (nearly) optimal solution for the originaldiscrete problem. Various rounding methods have been proposed in the last twenty five years. In this talk I will intro-duce a new probabilistic technique for transforming the fractional solution to an integral one so that the underlyingcombinatorial structure of the problem is preserved. The technique is based on sampling from maximum entropydistributions over combinatorial structures hidden in such problems. In order to present the idea and provide thehigh-level intuition behind it, I will go through the generalization of the Traveling Salesman Problem (AsymmetricTSP) and show how we can improve the worst-case performance guarantee for this problem after almost 30 years. Wewill also see other applications of this technique in assignment problems and fair resource allocation.

Speaker : Negar Soheili Azad (Tepper, Carnegie Mellon University, nsoheili@andrew.cmu.edu)Title : A Deterministic Rescaled Perceptron AlgorithmAbstract : The classical perceptron algorithm is a separation-based algorithm for solving conic-convex feasibility problems in

a number of iterations that is bounded by the reciprocal of the square of the cone thickness. We propose a modi-fied perceptron algorithm that leverages periodic rescaling for exponentially faster convergence, where the iterationbound is proportional to the logarithm of the reciprocal of the cone thickness and another factor that is polynomialin the problem dimension.

Coauthor(s) : Javier Pena (jfp@andrew.cmu.edu)

Speaker : Onur Babat (Lehigh University, onur.babat@lehigh.edu)Title : Linear solution scheme for the cardinality constrained portfolio allocation modelsAbstract : The cardinality-constrained portfolio allocation models can be formulated as a MIQP when the risk is measured by

the portfolio return variance. Thus it can be difficult to solve especially for the large scale instances. To tackle thisinherent difficulty, we propose a linear solution scheme based on Benders reformulation. The effectiveness of thelinear solution schemes is illustrated by numerical experiments.

Coauthor(s) : Luis Zuluaga (luis.zuluaga@lehigh.edu)

14

Speaker : Xi Bai (Lehigh University, xib210@lehigh.edu)Title : Risk parity in portfolio selection: models and algorithmsAbstract : The risk parity optimization problem aims to find such portfolios for which the contributions of risk from all assets

are equally weighted. Portfolios constructed using risk parity approach are a compromise between two well-knowntechniques: minimum variance optimization approach (MVO) and equally weighted approach (EW). In this talk, wediscuss the problem of finding portfolios that satisfy risk parity of either individual assets or groups of assets. Wedescribe the set of all risk parity solutions by using convex optimization techniques over quadrants, and also proposean alternative nonconvex least-square model whose set of optimal solutions includes all risk parity solutions, andpropose a modified formulation which aims at selecting the most desirable risk parity solution (according to somecriteria). Furthermore, we propose an alternating linearization framework to solve this nonconvex model. Numericalexperiments indicate the effectiveness of our technique in terms of both speed and accuracy.

Coauthor(s) : Katya Scheinberg (katyas@lehigh.edu)

Speaker : Rimvydas Baltaduonis (Gettysburg College, rbaltadu@gettysburg.edu)Title : An Experimental Study of Complex-Offer Auctions: Payment Cost Minimization versus Offer Cost MinimizationAbstract : A Payment Cost Minimization auction has been proposed as an alternative to the Offer Cost Minimization auction

for use in wholesale electric power markets with an intention to lower procurement cost of electricity. Efficiency con-cerns are raised for this proposal assuming that the true production costs would be revealed to the auctioneer in acompetitive market. Using an experimental approach, I compare the performance of two auctions, controlling forthe level of unilateral market power. I find that neither auction results in allocations that correspond to the true costrevelation. Two auctions perform similarly in terms of procurement cost and efficiency. Surprisingly, consumer pricesin a competitive environment approach the prices of an environment with market power. It appears that the expectedinstitutional effects for procurement cost and efficiency are greatly dominated by the effects of anti-competitive be-havior due to the offer complexity and a cyclical nature of market demand.

Speaker : Soutir Bandyopadhyay (Lehigh University, sob210@lehigh.edu)Title : Multiresolution Gaussian Process Model for the Analysis of Large Spatial Data Sets.Abstract : The recent breakthroughs in Bayesian hierarchical models have added new classes of models for handling nonstation-

ary spatial data and indirect measurements of the spatial process. This development in spatial statistics is coincidentwith emerging challenges in the geosciences involving new types of observations and comparisons of data to complexnumerical models. For example, as attention in climate science shifts to understand the regional and local changesin future climate there is a need to analyze high resolution regional simulations from climate models and to comparethem to surface and remotely sensed observations at fine levels of details. These kinds of geoscience applicationsare characterized by large numbers of spatial locations and the application of standard spatial statistics techniquesis often not feasible or will take an unacceptably long time given typical computational resources. Moreover, geo-physical processes tend to be nonstationary over space and there is also the need to apply statistical methods that donot assume a constant spatial dependence across a region. In this work we develop a new statistical model that ad-dresses both of these features of geophysical data and so fills a gap in current statistical methodology. Our approachcombines the representation of a field using a multiresolution basis with statistical models for processes on a latticeand introduces sparsity into the computations in a way that does not compromise covariance models with large scalecorrelations and models with many degrees of freedom.

Speaker : Getachew K Befekadu (University of Notre Dame, gbefekadu1@nd.edu)Title : On the connection between the reliability of systems and the notion of invariance entropyAbstract : The purpose of this talk is to establish a connection between the problem of reliability (when there is an intermittent

control-input channel failure that may occur between actuators, controllers and/or sensors in the system) and thenotion of controlled-invariance entropy of a multi-channel system (with respect to a subset of control-input channelsand/or a class of control functions). We remark that such a connection could be used for assessing the reliability(or the vulnerability) of the system, when some of these control-input channels are compromised with an external"malicious" agent that may try to prevent the system from achieving more of its goal (such as from attaining invarianceof a given compact state and/or output subspace).

Coauthor(s) : Vijay Gupta (vgupta2@nd.edu) , Panos J. Antsaklis (Antsaklis.1@nd.edu)

15

Speaker : Hande Y. Benson (Drexel University, benson@drexel.edu)Title : Interior-Point Methods within a MINLP FrameworkAbstract : In this talk, we present details of MILANO (Mixed-Integer Linear and Nonlinear Optimizer), a Matlab-based tool-

box for solving mixed-integer optimization problems. For MINLP, it includes implementations of branch-and-boundand outer approximation algorithms and solves the nonlinear subproblems using an interior-point penalty methodintroduced by Benson and Shanno (2005) for warmstarts. Special consideration is given for problems with cone con-straints.

Speaker : Xiaoqiang Cai (The Chinese University of Hong Kong, xqcai@se.cuhk.edu.hk)Title : Optimal Dynamic Stochastic Scheduling with Partial Losses of WorkAbstract : Stochastic scheduling subject to random machine breakdowns has been the focus of study in an extensive literature

over decades. Prior research in this area has been conducted, nevertheless, with the implicit assumption that a ma-chine breakdown causes either no loss of the work achieved on a job at all, or a total loss of the work achieved. In manypractical problems, however, the work achieved on a job is neither fully preserved nor totally lost after a breakdown.In this article, we develop a unified approach to deal with any level of partial losses due to machine breakdowns. Morespecifically, we consider a problem to process a number of jobs with arbitrary random processing times by a machinesubject to general stochastic breakdowns, where each breakdown may cause an uncertain loss of the work achievedon the job being processed. The objective is to maximize the expected weighted discounted reward of completingthe jobs. We develop a general framework to model uncertain losses of work caused by breakdowns as a semi-Markovtransition process. We obtain the optimal dynamic polices using multi-armed bandit process methodology, which arecharacterized by a set of Gittins indices as solutions to a system of integral equations.

Coauthor(s) : Xianyi Wu (xywu@stat.ecnu.edu.cn) , Xian Zhou (xian.zhou@mq.edu.au)

Speaker : Elcin Cetinkaya (Lehigh University , elcin.cetinkaya@gmail.com)Title : Portfolio Risk Management with Moment Matching ApproachAbstract : We investigate the problem minimizing the probability of obtaining a portfolio return less than a threshold while

keeping the expected portfolio return no worse than a target. We propose a tractable solution involving an algorithmbased on log-Normally distributed stock returns assumption and the Fenton-Wilkinson approximation method to theproblem difficult to solve using exact methods. We compare its performance to that of some benchmark methods. Weextend our approach to design basket options.

Coauthor(s) : Aurelie Thiele (aurelie.thiele@gmail.com)

Speaker : Chen Chen (Lehigh University, morningchen1983@gmail.com)Title : MPC-based Appliance Scheduling for Residential Building Energy Management ControllerAbstract : With the emerging smart grid enabling two-way communication, customers will be able to receive time-varying prices

of electricity; the price variations will in turn serve as incentives for customers to alter their power usage profiles. Manyresidential appliances, e.g., clothes washer/dryer and plug-in electric vehicle (PEV), provide operational flexibilitiesthat customers can exploit to take advantage of these pricing incentives. This flexibility can simultaneously benefitelectric utilities and grid operators by relieving peak demand. However, current residential load control activitiesare mainly operated manually, which poses great challenges to customers seeking to optimally schedule applianceoperations in the presence of time-varying electricity prices. Hence, an automated building energy managementcontroller (BEMC) is necessary to optimize the appliances’ operation on behalf of customers.In this work, we propose an optimization method for the BEMC to schedule appliances within buildings. Boththermostatically-controlled appliances (e.g., electric heaters) and non-thermal appliances (e.g., dishwasher, clotheswasher/dryer, PEVs) with flexibilities are considered in the proposed method. For non-thermal appliance scheduling,in which delay and/or power consumption flexibilities are available, we model the BEMC operations as a mixed-integer linear programming (MILP) problem, where operation dependence of inter-appliance and intra-appliance isintegrated to further exploit electricity price variations. For thermal appliance scheduling, the thermal mass of thebuilding, which serves as thermal storage, is integrated into the linear programming problem by modeling the ther-modynamics of rooms in a building as constraints. Within the comfort range modeled by the predicted mean vote(PMV) index, thermal appliances are scheduled smartly together with thermal mass storage to hedge against highprices and exploit low-price periods. To cope with uncertainty of prices and weather information, model predic-tive control (MPC) method, which incorporates both forecasts and newly updated information, is utilized to build arolling-based finite-horizon optimization. Simulation results show that customers have notable energy cost savingson their electricity bills when using the proposed BEMC optimization in the presence of time-varying prices.

Coauthor(s) : Shalinee Kishore (skishore@lehigh.edu) , Jianhui Wang (jianhui.wang@anl.gov) , Yeonsook Heo (yheo@anl.gov)

16

Speaker : Boris Defourny (Princeton University, defourny@princeton.edu)Title : A Nested Look-Ahead Model for Unit Commitment with Joint Ramping Capability RequirementsAbstract : We propose a formulation of the unit commitment problem where joint ramping capability requirement constraints

over multiple durations are introduced to help mitigating the risk posed by poorly predicted variations of wind energysupply over multiple time scales. The optimization model implements short-horizon look-aheads inside a main look-ahead over the planning horizon.

Coauthor(s) : Hugo P. Simao (hpsimao@princeton.edu) , Warren B. Powell (powell@princeton.edu)

Speaker : Jonathan Donadee (Carnegie Mellon University, jdonadee@andrew.cmu.edu)Title : Co-Optimization of Grid-to-Vehicle Charging and Ancillary ServicesAbstract : This talk investigates the charging of an electric vehicle (EV) that has access to both energy and ancillary services

markets. The EV’s decision optimization problem is formulated as a finite horizon MDP with multiple sources ofMarkovian uncertainty. The MDP is solved using a novel heuristic backwards recursion. This heuristic leverages linearprogramming theory to create piecewise-linear convex value functions and enables optimization over a continuousspace of actions.

Coauthor(s) : Marija Ilic (milic@ece.cmu.edu)

Speaker : Hongbo Dong (University of Wisconsin-Madison, hdong6@wisc.edu)Title : Conic relaxations for convex quadratic optimization with indicator variablesAbstract : We consider the problem of globally solving convex quadratic optimization with indicator variables, which has appli-

cations in portfolio optimization, sparse filter design, etc. We construct multiple convex conic relaxations, with spe-cial focus on second order cone programming and sparse semidefinite programming. Together with valid inequalitiesgenerated by exploiting relevant convex hulls in small dimensions, we evaluate the strength-complexity trade-offs ofthese convex relaxations.

Coauthor(s) : Jeff Linderoth (linderoth@wisc.edu)

Speaker : Yang Dong (Lehigh University, yad210@lehigh.edu)Title : Robust Manager Allocation for Investment ManagementAbstract : A key difficulty for an investment manager is to quantify fund managers’ skill when he may not know managers’

allocation precisely. We propose a portfolio optimization framework that takes into account the uncertainty in fundallocation and consider both a robust optimization approach and a stochastic perspective. We then provide insightsinto the impact of the uncertainty related to the asset allocation and the asset returns on portfolio performance andthe manager selection policy.

Coauthor(s) : Aurelie Thiele (aut204@lehigh.edu)

Speaker : Xingyuan Fang (Princeton University, xingyuan@princeton.edu)Title : Online PRSMAbstract : Online optimization has been shown to be powerful for large scale problem. We present a new online optimization

method Online PRSM which applies to a wide range of applications. An ergodic O(1/�

t ) convergence rate in expec-tation is also derived.

Speaker : Farbod Farhadi (University of Massachusetts Amherst, ffarhadi@som.umass.edu)Title : Column Generation and Accelerating Schemes for Mixed-Mode Aircraft Sequencing ProblemsAbstract : We discuss and contrast alternative accelerations schemes for a column generation approach with application to

mixed-mode aircraft sequencing problems. Such enhancements include solving the pricing sub-problem with dy-namic programming vs. MIP along with stabilization strategies. Computational results are reported.

Coauthor(s) : Ahmed Ghoniem (aghoniem@isenberg.umass.edu)

17

Speaker : Hanan Mostaghimi Ghomi (University of Windsor, )Title : Three Dimensional Knapsack Problem with Vertical Stability and Pre-Placed BoxesAbstract : A three-dimensional knapsack problem packs a subset of rectangular boxes inside a bin with fixed size such that the

utilization of the bin’s volume is maximized. Each box has its own value and size and can be freely rotated into anyof the six positions while its edges are parallel to the bin’s edges. A knapsack algorithm based on Eglebad & Pisinger(2009) sequence triple representation is developed. The sequence triple shows the relative position where the boxesmust be packed. The proposed algorithm considers box rotation and vertical stability. Simulated annealing techniqueand FFD algorithm are used to model the problem. Moreover, the situation where some boxes are pre-placed in the binis investigated. These pre-placed boxes represent potential obstacles. Numerical experiments have been conductedfor bins with and without obstacles. The initial results provide bin’s utilization of about 87%. More experiments arestill conducted to refine the results.

Coauthor(s) : Walid Abdul-Kader (kader@uwindsor.ca) , Hanan Mostaghimi (mostaghh@uwindsor.ca)

Speaker : Julio Góez (Lehigh University, jcg207@lehigh.edu)Title : Full characterization of disjunctive-conic-cuts for mixed integer second order cone optimizationAbstract : Mixed integer second order cone optimization (MISOCO) problems have an increasing number of engineering appli-

cations including supply chain, finance, and networks design. In this talk we analyze the derivation of Disjunctive-Conic-Cuts (DCCs) for MISOCO problems. We present a full characterization of the DCCs when the disjunctive setconsidered is defined by parallel hyperplanes. We also present the results of some preliminary computational experi-ences with the novel DCCs.

Speaker : Walter Gomez (Universidad de La Frontera, wgomez@ufro.cl)Title : Consumer demand systems based on discrete-continuous modelsAbstract : We consider the problem of describing consumer choice situations characterized by the simultaneous demand for

multiple alternatives that are imperfectly substitutes for one another. The econometric technique to deal with thisproblem is the so called Kuhn-Tucker multiple discrete-continuous economic consumer demand model. This modelis usually stated with suitable nonsymmetric error distribution such that closed forms for the underlying probabilityfunction of consumption patterns can be obtained.In this paper we modify the model in two ways. On the one we use another probability distribution for the errorthat is symmetric and also provides closed forms. On the other hand, we include semidefinite constraints into theoptimization problem arising by doing statistical inference with the model. Numerical experiments with the newmodel are presented.

Coauthor(s) : Ignacio Vidal (ividal@inst-mat.utalca.cl) , Felipe Vasquez (fvasquez@udec.cl) , Alejandro Llanos(allanos@ufro.cl)

Speaker : Clovis C. Gonzaga (Federal University of Santa Catarina, Brazil, ccgonzaga1@gmail.com)Title : On the complexity of steepest descent methods for minimizing convex quadratic functionsAbstract : We discuss the performance of the steepest descent algorithm for minimizing a quadratic function with hessian matrix

eigenvalues between 1/C and 1. Steepest descent methods differ exclusively on the choice of the step length at eachiteration. We develop a scheme for choosing the step lengths with the following result: the number K of iterationsneeded to reduce the objective function, the gradient norm and the distance to the optimal solution by a factor � isbounded by K ≤

�C log(1/�). This is contrasted to the linear dependence predicted by Kantorovich’s analysis.

18

Speaker : Chaitra Gopalappa (Futures Institute, cgopalappa@fututresinstitute.org)Title : Simulation model for the analyses and cost estimates of combination HIV-prevention strategies for the elimina-

tion of HIVAbstract : UNAIDS aims at achieving zero new HIV infections, zero HIV-related deaths, and zero discrimination. We analyze

alternative combinations of HIV-intervention programs to identify strategies that can achieve the goals on new in-fections and deaths and estimate the corresponding costs of the strategies. These programs, along with current keyHIV-prevention programs such as prevention of mother to child transmission and treatment with antiretroviral ther-apy (ART), also include new technologies such as ART as a prevention strategy to reduce risks of transmission (test andtreat), pre-exposure prophylaxis to reduce risk of acquisition of infection, and the prospective role of an effective HIVvaccine if it becomes available. We used a compartmental model, Goals, to simulate the strategies for 24 countriesthat contribute to 85% of new infections in low and middle income countries. We scaled-up the results to estimatethe total resources needed for 139 low and middle income countries. Estimates indicate that only a combination of allprograms, current and new technologies, can lead to a sustainable reduction in new infections and deaths and hencewill be critical to achieving the UNAIDS targets.

Coauthor(s) : John Stover , Carel Pretorius (on behalf of the New Prevention Technology Study Group)

Speaker : Jackie Griffin (Northeastern University, ja.griffin@neu.edu)Title : Patient-bed Assignments in Hospital SystemsAbstract : To alleviate overcrowding in hospitals, hospitals may implement policies that address the management of patient ar-

rivals through the redirection of patients to other hospitals. We model the hospital unit as a Markov chain and developtype-specific threshold policies for patient assignment while simultaneously addressing three distinct objectives.

Coauthor(s) : Pinar Keskinocak (pinar@isye.gatech.edu)

Speaker : Bin Han (University of Maryland, College Park, danielh@math.umd.edu)Title : Efficient Learning of Donor Retention Strategies for the American Red CrossAbstract : We present a new sequential decision model for adaptively allocating a fundraising campaign budget for a non-profit

organization such as the American Red Cross. The campaign outcome is related to a set of design features using linearregression. We derive the first simulation allocation procedure for simultaneously learning unknown regression pa-rameters and unknown sampling noise. The large number of alternatives in this problem makes it difficult to evaluatethe value of information. We apply convex approximation with a quantization procedure and derive a semidefiniteprogramming relaxation to reduce the computational complexity. Simulation experiments based on historical datademonstrate the efficient performance of the approximation.

Coauthor(s) : Ilya Ryzhov (iryzhov@rhsmith.umd.edu) , Boris Defourny (defourny@princeton.edu)

Speaker : Zheng Han (Lehigh University, zhh210@lehigh.edu)Title : A primal-dual active-set algorithm for large-scale convex quadratic optimizationAbstract : Active-set methods enjoy great popularity in practical nonlinear optimization problems despite the possible draw-

back of slow update of the active-set estimate. We present a primal-dual active-set framework for large-scale convexquadratic optimization that can make multiple simultaneous changes in the active-set estimate and converge from ar-bitrary initial points. The iterates of our framework are the active-set estimates themselves, where with each estimatea primal-dual solution is uniquely defined via a reduced subproblem. The computational cost of each subproblem istypically only modestly more than solving a reduced linear system. Moreover, we can potentially undercut further thecost of solving the subproblem by incorporating inexactness. Preliminary numerical results illustrate that our methodis efficient and can often lead to rapid identification of the optimal active-set even for poorly conditioned problems.

Coauthor(s) : Frank E. Curtis (frank.e.curtis@lehigh.edu) , Daniel P. Robinson (daniel.p.robinson@jhu.edu)

Speaker : Lin He (Lehigh University, lih308@lehigh.edu)Title : Inventory Management for a Distribution System Subject to Supply DisruptionsAbstract : We study inventory optimization for continuous-review multi-echelon distribution systems subject to supply dis-

ruptions, with Poisson customer demands under a first-come, first-served allocation policy. We develop a recursiveoptimization heuristic, which applies a bottom-up approach that sequentially approximates the base-stock levels ofall the locations. A preliminary numerical study shows that it performs very well, reaching within 1% of optimal formost instances tested.

Coauthor(s) : Larry Snyder (lvs2@lehigh.edu)

19

Speaker : Jason Hicken (Rensselaer Polytechnic Institute, hickej2@rpi.edu)Title : A Flexible Iterative Trust-Region Algorithm for Nonstationary PreconditionersAbstract : Matrix-free optimization algorithms are typically based on Krylov iterative methods that, in principle, require only

matrix-vector products. Such algorithms are attractive for reduced-space PDE-constrained optimization, where theHessian and Jacobian are computationally expensive to form explicitly. In practice, Krylov solvers also require pre-conditioning, especially for large-scale problems with many degrees of freedom. Quasi-Newton preconditioners havebeen used successfully in several applications, but this class of preconditioner may require many iterations before itis effective. To remain truly matrix-free, one possibility is to use a nested (inner) iterative solver as a preconditioner.Iterative solvers are nonstationary, in general, so we need to consider so-called flexible outer methods. Flexible Krylovmethods present unique challenges in the context of optimization, because they destroy the symmetry present in thesystem. In this work, we present a flexible iterative trust-region method based on solving the trust-region subproblemin a projected subspace. The proposed algorithm uses an ad hoc strategy to resolve the conflict between precondi-tioning the linear system and estimating the largest negative eigenvalue of the Hessian. Numerical experiments verifythe method and demonstrate its effectiveness on a PDE-constrained optimization problem.

Speaker : Hassan Lionel Hijazi (NICTA, hassan.hijazi@nicta.com.au)Title : Convex Quadratic Approximations of AC Power FlowsAbstract : New convex quadratic approximations of the AC power flow equations are studied. The approximation is motivated

by hybrid discrete/continuous applications in power systems that operate outside normal conditions. Under suchcircumstances, which arise in power restoration and renewable energy integration, existing approximations are inac-curate to be useful in practice. The convex quadratic approximations remedy these limitations by capturing reactivepower and voltage magnitude accurately. Two case studies in optimal power flows and capacitor placement, demon-strate the benefits of the new formulations in terms of accuracy and efficiency.

Coauthor(s) : Carleton Coffrin (Carleton.Coffrin@nicta.com.au) , Pascal Van Hentenryck (pvh@nicta.com.au)

Speaker : Robert Howley (Lehigh University, roh210@lehigh.edu)Title : Computing semiparametric bounds on the expected payments of insurance instruments via column generationAbstract : It has been recently shown that numerical semiparametric bounds on the expected payoff of financial or actuarial in-

struments can be computed using semidefinite programming. However, this approach has practical limitations. Herewe use column generation, a classical optimization technique, to address these limitations. From column generation,it follows that practical univariate semiparametric bounds can be found by solving a series of linear programs. In ad-dition to moment information, the column generation approach allows the inclusion of extra information about therandom variable; for instance, unimodality and continuity, as well as the construction of corresponding worst/best-case distributions in a simple way.

Coauthor(s) : Bob Storer (rhs2@lehigh.edu) , Luis Zuluaga (luis.zuluaga@lehigh.edu) , Juan Vera (j.c.veralizcano@uvt.nl)

Speaker : Cho-Jui Hsieh (UT Austin, cjhsieh@cs.utexas.edu)Title : Sparse Inverse Covariance Matrix Estimation Using Quadratic ApproximationAbstract : The L1-regularized Gaussian maximum likelihood estimator (MLE) has been shown to have strong statistical guar-

antees in recovering a sparse inverse covariance matrix, or alternatively the underlying graph structure of a GaussianMarkov Random Field, from very limited samples. We propose a novel algorithm for solving the resulting optimiza-tion problem which is a regularized log-determinant program. In contrast to recent state-of-the-art methods thatlargely use first order gradient information, our algorithm is based on Newton’s method and employs a quadratic ap-proximation, but with some modifications that leverage the structure of the sparse Gaussian MLE problem. We showthat our method is superlinearly convergent, and present experimental results using synthetic and real-world appli-cation data that demonstrate the considerable improvements in performance of our method when compared to otherstate-of-the-art methods.

Coauthor(s) : Matyas Sustik (sustik@cs.utexas.edu) , Inderjit Dhillon (inderjit@cs.utexas.edu) , Pradeep Ravikumar(pradeepr@cs.utexas.edu)

20

Speaker : Brian J. Hunt (Johnson C. Smith University, bjhunt@jcsu.edu)Title : A Modeling and Simulation Approach to Emergency ManagementAbstract : We present a model of North Carolina Emergency Management’s disaster logistics process for fulfilling disaster supply

requests from state and local agencies. The model will provide an advance planning tool to determine the optimumlogistics process resources (request processors, supply purchasers, delivery trucks, etc.) required to meet performanceobjectives for delivery depending on predicted daily demand levels. Data collected during the responses to HurricanesIrene (Category 1, 2011) and Floyd (Category 3, 1999) was used to develop the model.

Coauthor(s) : Paul Latham (paul.latham@ncdps.gov)

Speaker : Jhi-Young Joo (Carnegie Mellon University, jjoo@ece.cmu.edu)Title : Adaptive Load Management: Scheduling And Coordination Of Demand Resources In Power SystemsAbstract : Demand response refers to techniques that manage end-users’ electricity consumption in order to help the power

system operate in a more cost-efficient and reliable way. It is becoming more important with increasing renewableand distributed energy resources incorporated into the system. Our proposed demand response framework, namelyAdaptive Load Management, provides a comprehensive structure for demand response by formulating the complexpower system objective as many sub-problems of diverse supply and demand entities in the system over multiple timehorizons. In this talk, we will focus on the short-term scheduling of supply and demand resources, with an emphasison managing flexible demand resources of small end-users. The difficulty of this problem comes from the uncertaintyof loads and supply, limitations on communication and information exchange among a large number of supply anddemand entities, and modeling different values of electric energy seen by diverse end-users/loads. We will addresshow we tackle these issues, and present simulation results that demonstrate how our methodology works.

Coauthor(s) : Marija Ilic (milic@ece.cmu.edu)

Speaker : Abdullah Konak (Penn State Berks, konak@psu.edu)Title : Cyclic Facility Layout Problem: A Hybrid Exact/Heuristic Optimization ApproachAbstract : This presentation introduces a special case of Dynamic Facility Layout Problem (DFLP) where product types, product

demands, and departmental area requirements are seasonal. There are several production periods in each planninghorizon, and the production periods repeat themselves in consecutive planning horizons. Therefore, this new prob-lem is called Cyclic Facility Layout Problem (CFLP). In this study, first, a mixed integer programming formulationis introduced for the CFLP. The proposed formulation relaxes the assumption of fixed department shapes, which iscommonly accepted in the DFLP literature. In addition, department sizes are allowed to change over the planninghorizon. This relaxation is required particularly in real-world cases where the area requirements of departments arealso seasonal, and the facility size is limited. Then, a large scale hybrid simulated annealing (LS-HSA) is proposed tosolve the proposed formulation for problem instances with practical sizes. The proposed LS-HSA has shown to beeffective and versatile as it can be applied to various facility layout problems.

Coauthor(s) : Sadan Kulturel-Konak (sadan@psu.edu)

Speaker : Sadan Kulturel-Konak (Penn State Berks, sadan@psu.edu)Title : Solving the Unequal Area Facility Layout Problem: An Effective Hybrid Optimization Strategy Coupled with the

Location/Shape RepresentationAbstract : In this study, an innovative integration of Genetic Algorithms (GA) and Linear Programming (LP) is proposed to solve

the Facility Layout Problem (FLP) on a continuous plane with unequal-area departments. In addition, a new encod-ing scheme, called the location/shape representation, has been developed. The proposed hybrid LP/GA approach isdifferent from previously reported hybrid heuristic/exact algorithms in several ways. LP is not only used to evaluatesolutions represented in the GA’s encoding scheme but also, the output of the LP solution is directly integrated into theGA’s encoding. The GA generates new solutions by recombining prior solutions obtained by the LP. This integrationof the output of the LP into the GA’s encoding is enabled by the location/shape representation. The superior perfor-mance of the proposed hybrid GA/LP approach can be attributed to the location/shape representation and how theinformation gained by the LP is integrated into the GA’s encoding using the location/shape representation.

Coauthor(s) : Abdullah Konak (konak@psu.edu)

21

Speaker : Andrew Lambe (University of Toronto, lambe@utias.utoronto.ca)Title : A Matrix-Free Augmented Lagrangian Algorithm for Large-Scale Structural DesignAbstract : This talk considers the problem of minimum-mass structural design subject to stress constraints. Problems in this

class include not only PDE constraints, but also a large number of general (i.e. maximum stress) constraints. Manystructures found in aircraft design problems need to be modeled with high-order finite element methods to accuratelypredict their behavior. However, in the reduced-space form of the problem, the cost of computing the gradients toall the stress constraints dominates the computational cost. We present an augmented Lagrangian optimization al-gorithm for solving this reduced-space problem that is matrix-free, i.e. the algorithm requires neither the full Hessiannor full Jacobian of the optimization problem, only appropriate matrix-vector products. Test results show that thismatrix-free approach is competitive with an SQP algorithm (which requires the full Jacobian) on the reduced-spaceproblem and scales well with refinement of the finite element mesh. We also comment on extensions of this work tosolving full-space PDE-constrained problems.

Coauthor(s) : Sylvain Arreckx (sylvain.arreckx@polymtl.ca) , Joaquim Martins (jrram@umich.edu) , Dominique Orban(dominique.orban@gerad.ca)

Speaker : Eric Landquist and Francis Vasko (Kutztown University, elandqui@kutztown.edu)Title : A Simple and Efficient Strategy for Solving Large Generalized Cable-Trench ProblemsAbstract : Vasko et al. (2002) defined the Cable-Trench Problem (CTP) as a combination of the Shortest Path and Minimum

Spanning Tree Problems. Specifically, let G = (V ,E) be a connected weighted graph with specified vertex v1 ∈ V (re-ferred to as the root node), weight l (e) ≥ 0 for each e ∈ E , and positive parameters τ and γ. The Cable-Trench Problem(CTP) is the problem of finding a spanning tree T of G such that τlτ(T )+γlγ(T ) is minimized, where lτ(T ) is the totallength of the spanning tree T and lγ(T ) is the total path length in T from v1 to all other vertices of V . Recently, Jiang etal. (2011) modeled the vascular connectivity problem in medical image analysis as a large generalization of the CTP.They proposed an efficient solution based on a modification of Prim’s algorithm (MOD_PRIM), but did not elaborateon it. In this paper, we will formally define the Generalized CTP (GCTP) and describe MOD_PRIM in detail. We showthat MOD_PRIM gives nearly optimal results for small GCTPs and describe two heuristics which further improve theresults of MOD_PRIM. These algorithms are capable of finding nearly optimal solutions of very large GCTPs as effi-ciently as theoretically possible. Empirical results for graphs with up to 25,000 vertices and about 11 million edges willbe given.

Coauthor(s) : Francis Vasko (vasko@kutztown.edu) , Gregory Kresge (gkres121@live.kutztown.edu) , Adam Tal(atal822@live.kutztown.edu) , Yifeng Jiang (jiang1feng@gmail.com) , Xenophon Papademetris(xenophon.papademetris@yale.edu)

Speaker : Yanchao Liu (University of Wisconsin-Madison, yliu67@wisc.edu)Title : Modeling Demand Response for FERC Order 745Abstract : We study the FERC Order 745 regarding demand response compensation in organized wholesale energy markets, and

explore different approaches to model and solve a compliant implementation of the Order. In the economics sense,demand response in the Order context is a trade of “consuming right” instead of a trade of energy, therefore it mustbe dispatched separately from the economic dispatch of energy. This dictates that in general, simultaneous clearingof demand response and energy can only be achieved in one of the two ways: an iterative process, or a hierarchicalmodel. We find a bi-level optimization model to be a suitable one for the given problem. The lower level performsthe economic dispatch of energy and generates the price, and the upper level minimizes the total compensation fordemand response subject to the net benefit requirement. We show that the con-convex net benefit test requirementcan be transformed to a linear constraint in the bi-level framework, which greatly eases the computational complexity.Experiments show that the solution process is reliable and significantly outperforms, in terms of accuracy and speed,a heuristic algorithm implementing the iterative approach. We apply the model to various data cases and settings,and generate useful insight for dispatch operations in compliance with the Order.

Coauthor(s) : Michael C. Ferris (ferris@cs.wisc.edu)

22

Speaker : Roger Lueken (Carnegie Mellon University, rlueken@andrew.cmu.edu)Title : The effects of bulk electricity storage on the PJM marketAbstract : In this talk, we will present a unit commitment and economic dispatch (UCED) model of the PJM Interconnection.

This model is a mixed integer linear optimization that minimizes the total operational costs of providing electricity.Similar UCED models are used to manage all major restructured electricity markets. Our model, entitled PHORUM, isopen source and is designed to be easily used and improved upon by other researchers interested in electricity policy.We will next discuss an application of PHORUM: analyzing the effect of bulk electricity storage on grid operations. Intoday’s electricity industry, storing large quantities of power is economically infeasible. Therefore, total generationmust match demand at all times. This inefficiency prevents the electric sector from operating as a conventional com-petitive market that relies on inventory. However, new affordable grid-scale storage technologies such as advancedbatteries are emerging that will allow supply and demand to decouple, fundamentally altering how the electric systemand its markets are operated. Our research analyzes how grid operations change when large amounts of storage areavailable. We are interested in how this deployment will affect the following: Emissions of CO2, and other pollutants;Electricity prices for consumers; Profits of generators; Profits of storage operators.

Coauthor(s) : Jay Apt (apt@cmu.edu)

Speaker : Biao Mao (Rensselaer Polytechnic Insitute, maob@rpi.edu)Title : Environmental SuperOPF Electricity Market Planning ToolAbstract : There are quite wide ranges of optimization problems in the power system and electricity market. Optimization prob-

lems dealing with the short run include unit dispatch and operation of markets. Optimization problems dealing withthe long run include design of markets and optimal investment. Furthermore renewable energy generation will be in-creasingly integrated in the power system in the future so it is also very important to evaluate the impacts of changesin technology as well as environmental regulations promoting them. Instead of breaking down all the relevant opti-mization problems into several sub-problems, our “SuperOPF Planning Tool” is a framework which integrates engi-neering, economic and environmental models of the power system through co-optimization methods across multi-ple scenarios. The foundation of the SuperOPF is the extensible optimal power flow formulation according to laws ofphysics. Then it predicts and optimizes system operation, investments, and retirements through a multi-scenario co-optimization approach, which also includes environmental impacts such as emissions and estimated health damage.One of the activities we can do through all of these models and methods is estimate the long-run effects of environ-mental policies on the electricity grid. We develop a model of power grid in the eastern US and Canada which includesthe transmission lines, locational demand functions, the grid operator’s decision-making, and locational marginalprices. We then use these to predict the impact of a carbon dioxide cap-and-trade program, a continuation of thecurrent subsidies for wind and solar power generators, a ban on new coal-burning power plants, and higher-than-expected natural gas prices. In order to predict the impacts of these policy and price conditions in the long run, wesimulate the model over three time periods. The base year is 2012, while investment is allowed in the decades endingin 2022 and 2032. The results of these simulations show that this model can predict optimal operation, investment,retirement, prices, economic surplus, emissions, and health effects of policies that affect the electric power industry.

Coauthor(s) : Daniel L. Shawhan (shawhd@rpi.edu) , John T. Taber (jtt24@cornell.edu) , Ray D. Zimmerman(rz10@cornell.edu) , Di Shi , Daniel Tylavsky (tylavsky@asu.edu) , Charles M. Marquet (cmm363@cornell.edu) ,Jubo Yan (jy489@cornell.edu) , Richard E. Schuler , William D. Schulze

Speaker : Nikola Markovic (University of Maryland, College Park, nikola.didi@gmail.com)Title : Evasive Flow Capture: Optimal Location of Weigh-in-Motion Systems, Tollbooths, and Safety CheckpointsAbstract : The flow-capturing location-allocation problem (FCLAP) consists of locating facilities in order to maximize the num-

ber of flow-based customers that encounter at least one of these facilities along their predetermined travel paths. InFCLAP, it is assumed that if a facility is located along (or “close enough” to) a predetermined path of a flow, the flow ofcustomers is considered captured. However, existing models for FCLAP do not consider the likelihood that targetedusers may exhibit non-cooperative behavior by changing their travel paths to avoid fixed facilities. Examples of facil-ities that targeted subjects may have an incentive to avoid include weigh-in-motion stations used to detect and fineoverweight trucks, tollbooths, and security and safety checkpoints. This paper introduces a new type of flow capturingmodel, called the “Evasive Flow Capturing Problem” (EFCP), which generalizes the FCLAP and has relevant applica-tions in transportation, revenue management, and security and safety management. We discuss several variants ofEFCP and formulate a model for the optimal location of weigh-in-motion stations. We analyze structural propertiesof EFCP, propose exact and approximate solution techniques, and show an application to a real-world transportationnetwork.

Coauthor(s) : Ilya O. Ryzhov (iryzhov@rhsmith.umd.edu) , Paul Schonfeld (pschon@umd.edu)

23

Speaker : Tim Mitchell (Courant Institute of Mathematical Sciences, New York University, tim.mitchell@cims.nyu.edu)Title : A BFGS-based SQP Method for Constrained Nonsmooth, Nonconvex OptimizationAbstract : We consider constrained nonsmooth, nonconvex optimization problems, where both the objective and the con-

straints may be nonsmooth and nonconvex and are not assumed to have any special structure. In 2012, Curtis andOverton presented a gradient-sampling-based SQP algorithm with a steering strategy to control exact penalty penal-ization, proving convergence results that generalize the results of Burke, Lewis and Overton and Kiwiel for the uncon-strained problem. This algorithm uses BFGS approximation to define a “Hessian” matrix H that appears in the theQPs, but in order to obtain convergence results, upper and lower bounds on the eigenvalues of H must be enforced.On the other hand, Lewis and Overton have argued that in the unconstrained case, a simple BFGS method is muchmore efficient in practice than gradient sampling, although the Hessian approximation H typically becomes very illconditioned and no general convergence results are known. We consider an SQP method for the constrained problembased on BFGS approximation without gradient sampling, and ask the question: does allowing ill-conditioning in Hlead to the same desirable convergence behavior in practice as in the unconstrained case, or does the disadvantage ofsolving ill-conditioned QPs overcome any benefit gained by ill-conditioning? We test the algorithm on some simpleexamples as well as some challenging applied problems from feedback control.

Coauthor(s) : Frank E. Curtis (frank.e.curtis@lehigh.edu) , Michael L. Overton (overton@cs.nyu.edu)

Speaker : M.Mohsen Moarefdoost (Lehigh University, mom211@lehigh.edu)Title : Generation and Storage Dispatch in Electricity Networks with Generator DisruptionsAbstract : We present methods for optimizing generation and storage decisions in an electricity network with multiple unreliable

generators, each co-located with one storage unit, and multiple loads under power flow constraints. This problemcannot be optimized easily using stochastic programming and/or dynamic programming (DP) approaches. There-fore, in this study, we present several heuristic methods to find an approximate optimal solution for this system. Eachheuristic involves decomposing the network into several single- generator, single-battery, multi-load systems andsolving them optimally using dynamic programming, then obtaining a solution for the original problem by recom-bining. We discuss computational performance of the proposed heuristics as well as insights gained from the models.

Coauthor(s) : Lawrence V. Snyder

Speaker : Murat Mut (Lehigh University, mhm309@lehigh.edu)Title : A Tight Iteration-complexity Bound for IPM via Redundant Klee-Minty CubesAbstract : We consider two curvature integrals for the central path of a polyhedron. Redundant Klee-Minty cubes (Nematollahi

et al., 2007) have central path whose geometric curvature is exponential in the dimension of the cube. We prove ananalogous result for the curvature integral introduced by Sonnevend et al. 1990. Within an algorithmic framework,we rigorously prove that the iteration-complexity upper bound for the Klee-Minty cubes is tight.

Coauthor(s) : Tamás Terlaky (terlaky@lehigh.edu)

Speaker : Selvaprabu (Selva) Nadarajah (Tepper School of Business, Carnegie Mellon University, snadaraj@andrew.cmu.edu)Title : Relaxations of Approximate Linear Programs for the Real Option Management of Commodity StorageAbstract : The real option management of commodity conversion assets gives rise to intractable Markov decision processes

(MDPs). This intractability is due primarily to the high dimensionality of a commodity forward curve, which is partof the MDP state when using high dimensional models of the evolution of this curve, as commonly done in prac-tice. Focusing on commodity storage, we develop a novel approximate dynamic programming approach to obtainvalue function approximations from tractable relaxations of approximate linear programs (ALPs). We estimate lowerbounds and dual upper bounds on the value of an optimal policy on existing natural gas storage instances using thevalue function approximation from each of our models. Our ALP relaxations significantly outperform their corre-sponding ALPs in terms of both the estimated lower and upper bounds. Our approach is also relevant for the approxi-mate solution of MDPs that arise in the real option management of other commodity conversion assets, as well as thevaluation and management of real and financial options that depend on forward curve dynamics.

Coauthor(s) : Francois Margot (fmargot@andrew.cmu.edu) , Nicola Secomandi (ns7@andrew.cmu.edu)

24

Speaker : Daniel Olivares (University of Waterloo, dolivare@uwaterloo.ca)Title : A Centralized Energy Management System for Isolated MicrogridsAbstract : This work presents the mathematical formulation of the microgrid’s energy management problem and its implemen-

tation in a centralized Energy Management System (EMS) for isolated microgrids. Using model predictive control,the optimal operation of the microgrid is determined using an extended horizon of evaluation and recourse, whichallows proper dispatch of generators and energy storage units. The microgrid is modelled as a three-phase unbal-anced system with presence of both dispatchable and non-dispatchable distributed generation. In order to avoid amixed-integer non-linear formulation, the energy management problem is decomposed into a mixed-integer linearprogramming problem for Unit Commitment (UC), and a non-linear formulation for a multi-stage Optimal PowerFlow (OPF). A novel EMS architecture is proposed, which allows the multi-stage OPF to communicate reactive powerproblems to be corrected in the UC problem by committing additional capacity.The proposed EMS is tested in an isolated microgrid model based on a CIGRE medium-voltage benchmark system.Results justify the need for detailed three-phase models of the microgrid in order to properly account for voltage limitsand reactive power support.

Coauthor(s) : Claudio Canizares (ccanizares@uwaterloo.ca) , Mehrdad Kazerani (mkazeran@ecemail.uwaterloo.ca)

Speaker : Camilo Ortiz (Georgia Institute of Technology, camiort@gatech.edu)Title : An inexact block-decomposition CG hybrid method for dense and large-scale conic programmingAbstract : In this paper we introduce a new inexact two-block-decomposition first-order method for large-scale conic semidef-

inite programming. With a proper decomposition of the optimality conditions and the use of inexact error criteria inthe subproblems, this method permits the use of iterative methods like the conjugate gradient (CG) to approximatelyproject onto the manifold defined by the affine constraints. We prove the rate of convergence of the algorithm, takinginto account the error measures of the approximate solutions obtained from the CG method. As a result of allowinginexact projections onto the manifold, this method can solve problems of size and density that no other state-of-artblock-decomposition/split-operator method could solve before. In particular, we solve instances with more than fivemillion constraints and more than fifty million non-zero coefficients in the affine constraints.

Coauthor(s) : Renato D. C. Monteiro (monteiro@isye.gatech.edu) , Benar F. Svaiter (benar@impa.br)

Speaker : Haotian Pang (Princeton University, hpang@princeton.edu)Title : Estimating Sparse Precision Matrix by the Parametric Simplex MethodAbstract : We use a variant of simplex method called the parametric simplex method to estimate the sparse precision matrix

for some given data. The method is based on an important sparse precision matrix estimator called CLIME and aparametric simplex linear programming solver developed by us.

Coauthor(s) : Han Liu (hanliu@princeton.edu) , Robert Vanderbei (rvdb@princeton.edu)

25

Speaker : Dimitrios Papadimitriou (UGent & Bell Labs, dimitrios.papadimitriou@ugent.be)Title : Multi-agent information routing under dynamic and uncertain conditionsAbstract : In today’s communication networks, distributed control functions such as routing, are driven by memory consump-

tion, routing path quality properties (such as cost and length), adaptation cost, and convergence time. Direct formu-lations of the problem of determining an optimal routing obeying the protocol rules are complex to solve, and integerprogramming approaches can typically resolve only small to medium size routing problems [Bley2010]. Moreover, inlarge-scale information networks additional constraints appear (such as cache occupancy and content availability)that are difficult to integrate in a mixed-integer programming formulations. Therefore, heuristics are often necessaryto find good feasible solutions in a limited computing time. These heuristics also allow taking into account additionalrobustness requirement like, e.g., demand uncertainty [Altin2013]. To compensate the absence of mathematical mod-els for optimization, most approaches rely on simulation and scenario-based execution. Both exact and simulation-based approaches suffer however from two main drawbacks: 1) they only provide centralized optimization techniquescontradicting the intrinsic distributed computation nature of the decision problem and 2) they hardly take into ac-count uncertainties in the demand or the changes that arise in the underlying network such as link/node failures orunavailability of information servers. Note that uncertainty itself becomes a major element to consider for ensuringrobustness when designing large-scale information networks.Distributed Constraint Optimization Problem (DCOP) is a well-known attempt to perform distributed optimizationin a multi-agent framework, in which a group of agents must, in a distributed way, choose values for a set of variablessuch that the cost is either minimized or maximized and a set of constraints over the variables is satisfied (ADOPT[Pragnesh2005] and its improved version BnB-ADOPT [Yeoh2010]). More recently, attempts have been made to in-tegrate Lagrangean relaxation techniques [Gordon2012] and stochastic models [Nisan2012] in this framework. Onthe other hand, demand uncertainties can be modeled with the help of probabilistic distributions, although solvingthe resulting problem is difficult. Another more conservative approach consists in ensuring the good behavior inthe worst-case scenario. In both approaches, the resulting optimization model is often too large to be addressed byoff-the-shelve solvers.Consequently, the procedures and their associated parameters that drive the behavior of the distributed routing func-tion are designed in such a way that they arise naturally for the optimization models, allowing in turn for scalableand distributed procedures that approximate as much as possible the optimal solution of the global model. For thispurpose, we integrate decomposition-based methods for large scale optimization problem in the information routingfunctions. More precisely, we embed decomposition methods such as Benders, Dantzig-Wolfe or Lagrangean decom-position in DCOP (DCOP) in order for the distributed routing algorithm to perform as an asynchronous multi-agentsystem while the sub-problems appearing during the decomposition need to be solved very quickly. One of the mostimportant tasks when designing these models is to find a decomposition scheme that allows the sub-problems to besolved efficiently in a distributed way. Moreover, by design, re-optimization in the event of varying network conditionsand demands performs naturally in a smooth and distributed way.References:[Altin2013] A.Altin, B.Fortz, M. Thorup, and H.Umit, Intra-domain traffic engineering with shortest path routing pro-tocols, Annals of Operations Research, vol.204, pp.65Ð95, 2013.[Bley2010] A.Bley, B.Fortz, E.Gourdin, K.Holmberg, O.Klopfenstein, M.PiUro, A. Tomaszewski, and H.Emit. Optimiza-tion of OSPF Routing in IP Networks, In A. Koster and X. Munoz, editors, Graphs and Algorithms in CommunicationNetworks: Studies in Broadband, Optical, Wireless and Ad Hoc Networks, Ch.8, pp.199-240, Springer, 2010.[Gordon2012] G.J. Gordon, P.Varakantham, W.Yeoh, H.C.Lau, A.S.Aravamudhan, and S.-F.Cheng, Lagrangian relax-ation for large-scale multi-agent planning, Proc. of AAMAS 2012, pp.1227-1228, Valencia, Spain, June 2012.[Pragnesh2005] J.M.Pragnesh, W.-M.Shen, M.Tambe, and M.Yokoo, Adopt: asynchronous distributed constraint opti-mization with quality guarantees. Artificial Intelligence, vol.161(1-2), pp.149-180, 2005.[Yeoh2010] W.Yeoh, A.Felner, and S.Koenig, BnB-ADOPT: An Asynchronous Branch-and-Bound DCOP Algorithm.Journal of Artificial Intelligence Reseasrch (JAIR), vol.38, pp.85-133, 2010.

Coauthor(s) : Bernard Fortz (bernard.fortz@ulb.ac.be)

26

Speaker : Eugene Perevalov (Lehigh University, eup2@lehigh.edu)Title : On optimal information extraction from large-scale datasetsAbstract : The main promise of Big Data lies in the potential availability of information relevant for many practically important

problems. The main challenge is in the extraction of this information from the overwhelming volume of irrelevantdata. We consider a general information-theoretic framework for description of information accuracy and relevanceattributes (as opposed to just quantity) and its application to the Big Data challenge.

Coauthor(s) : David Grace (dpg3@lehigh.edu)

Speaker : Rommel G. Regis (Saint Joseph’s University, rregis@sju.edu)Title : Handling Equality Constraints in Expensive Black-Box Optimization Using Radial Basis Function SurrogatesAbstract : Surrogate models or metamodels have been used to approximate inequality constraints in expensive black-box opti-

mization. In particular, algorithms that utilize radial basis function (RBF) surrogates have been successfully appliedto a problem with over a hundred decision variables and with many black-box inequality constraints even when nofeasible starting points are given. However, not much work has been done on using surrogate models for optimiza-tion problems with expensive black-box equality constraints. This talk will present an RBF algorithm that can handleexpensive black-box equality constraints and provide numerical results on test problems.

Speaker : Ekkehard W. Sachs (University of Trier, sachs@uni-trier.de)Title : Preconditioners for PDE Constrained OptimizationAbstract : In this talk we consider preconditioning quadratic optimization problems which stem from optimization with partial

differential equations. We review some of the recent literature in this field. We propose an approach where a reducedorder model is used to design a preconditioner. This falls into the class of deflated Krylov space method. We presentnumerical results which show the efficiency of the approach.

Coauthor(s) : X. Ye (ye@uni-trier.de)

Speaker : Umit Saglam (Drexel University, us26@drexel.edu)Title : Multiperiod Portfolio Optimization with Cone Constraints and Discrete DecisionsAbstract : We consider a portfolio optimization problem where the investor’s objective is to choose a trading strategy that max-

imizes expected return penalized by transaction costs. We include portfolio diversification constraints in our singleand multiperiod models. The overall problem is a mixed-integer second-order cone programming problem, which wesolve with the Matlab-based solver MILANO. This talk will focus on the solution and warm-start of the second-ordercone programming subproblems.

Coauthor(s) : Hande Benson (hvb22@drexel.edu)

Speaker : Ali Kemal Sinop (IAS, asinop@cs.cmu.edu)Title : Approximation Algorithms for Graph Partitioning Problems using SDP HierarchiesAbstract : Graph partitioning is a fundamental optimization problem that has been intensively studied. Many graph partition-

ing formulations are important as building blocks for divide-and-conquer algorithms on graphs as well as to manyapplications such as VLSI layout, packet routing in distributed networks, clustering and image segmentation. Un-fortunately such problems are notorious for the huge gap between best known approximation algorithms and hard-ness of approximation results. In this talk, I will present an intuitive rounding algorithm for such problems usingLasserre/Parillo SDPs and relate the quality of its output to the spectral and isoperimetric profiles of the underlyinggraph.

Speaker : Lawrence V. Snyder (Lehigh University, lvs2@lehigh.edu)Title : Optimizing Locations for Wave Energy Farms under UncertaintyAbstract : We present models and algorithms for choosing optimal locations of ocean wave-energy conversion (WEC) devices

within a wave farm. The location problem can have a significant impact on the total power output of the farm due tothe hydrodynamic interactions between ocean waves and the waves created by the WECs themselves. The problem ishighly nonconvex and few rigorous optimization approaches have been proposed to solve it. Moreover, many authorshave lamented the fact that a wave farm optimized for a particular wave environment tends to perform quite poorlywhen the environment (e.g., wave angle or frequency) changes even slightly.In this talk, we introduce heuristics for solving both deterministic and stochastic or robust WEC location problems.We show that significantly more robust solutions than those in the literature can be obtained through stochastic orrobust optimization approaches, the first such demonstration of this fact. We also describe qualitative analysis suchas how the WEC locations change as the characteristics of the uncertain parameters, or the approach to uncertainty,changes.

Coauthor(s) : Lizhou Mao (lim311@lehigh.edu)

27

Speaker : Yunfei Song (Lehigh University, yus210@lehigh.edu)Title : Convex Sets as Invariant Sets for Linear SystemsAbstract : We propose a novel, unified, general approach to analyze sufficient and necessary conditions under which polyhedra,

polyhedral cones, ellipsoids, or Lorenz cones are invariant sets for a linear continuous or discrete system. We showthat the forward Euler method preserves positive invariance for polyhedra or polyhedral cones, while the backwardEuler method preserves positive invariance for all the four sets.

Coauthor(s) : Tamás Terlaky (terlaky@lehigh.edu)

Speaker : Arun Sridharan (Indian Institute of Technology Madras, Chennai, India, arunsriker@gmail.com)Title : Scheduling of multiproduct pipelines for transporting liquid fuelsAbstract : Petroleum products from refineries are transported to various distribution depots through multi-product pipelines,

road tankers, rail tankers and water carriers. Among them multi-product pipelines are gaining a lot of importanceand have several advantages over others. They carry various liquid fuels like gasoline, kerosene, diesel etc. that arepumped back-to-back as batches through the same pipeline from the refinery to various distribution depots and thento the marketing terminals to meet the overall customer demand. Between the product batches in the pipeline aninterface material called transmix is formed by mixing of products. These materials are either removed at the end ofthe pipeline or mixed with a lower grade of the same product. Drag reducing agents (DRA) is injected in the pipelinewhich reduces the frictional pressure drop thereby increasing the throughput and/or reducing the pumping cost.The challenge in multi-product pipeline operation is to select an optimum product sequence (volume and flow rateof product to be shipped) so as to minimize overall operating costs subject to constraints. The operating costs includethe pumping cost of products from refineries to distribution depots, the inventory cost of products at refinery anddistribution depots, cost of DRA, reprocessing cost of interface material and high-energy interval cost which resultsin a higher pumping cost when pipeline is operated during high-energy interval.In the literature, there are various approaches for solving scheduling problems. But the use of DRA in solving theproblem is not addressed in the literature. The proposed approach is based on improving the continuous-time Mixed-integer linear programming (MILP) formulation developed by Cafaro and CerdG (2004) by incorporating DRA com-ponent in the model that minimizes the overall operating costs subjected to various constraints. Constraints includeproduct-sequencing constraints, product-tracking constraints, product-inventory constraints etc. The model is ca-pable of assigning discrete levels of DRA concentration to various batches and thus achieving a reduction in pumpingcost. The resulting optimization problem is a MILP which is modelled in the ILOG/OPL framework and solved usingCPLEX.Cafaro, D. C., & Cerdá, J. (2004). Optimal scheduling of multiproduct pipeline systems using a non-discrete MILPformulation. Computers and Chemical Engineering, 28, 2053.

Coauthor(s) : Sridharakumar Narasimhan (sridharkrn@iitm.ac.in) , Shankar Narasimhan (naras@iitm.ac.in)

Speaker : Andy Sun (Georgia Institute of Technology, andy.sun@isye.gatech.edu)Title : Security-Constrained Optimal Power Flow with Sparsity Control and Efficient Parallel AlgorithmsAbstract : In this paper, we propose a new model for security-constrained DC optimal power flow problem, which can find a

post-contingency corrective action with a minimum number of adjustments. Our proposed formulation produces ageneration schedule, which has essentially the same low generation cost as the conventional corrective model, butrequires much fewer number of corrective actions. We also propose two decomposition algorithms to solve the prob-lem, which can be parallelized. Extensive computational results are presented for several standard IEEE test systemsand large-scale real-world power networks.

Coauthor(s) : Dzung T. Phan (phandu@us.ibm.com)

28

Speaker : Daniel B. Szyld (Temple University, szyld@temple.edu)Title : Inexact and truncated Parareal-in-time Krylov subspace methods for parabolic optimal control problemsAbstract : We study the use of inexact and truncated Krylov subspace methods for the solution of the linear systems arising in

the discretized solution of the optimal control of a parabolic partial differential equation. An all-at-once temporaldiscretization and a reduction approach are used to obtain a symmetric positive definite system for the control vari-ables only, where a Conjugate Gradient (CG) method can be used at the cost of the solution of two very large linearsystems in each iteration. We propose to use inexact Krylov subspace methods, in which the solution of the two largelinear systems are not solved exactly, and their approximate solutions can be progressively less exact. The option wepropose is the use of the parareal-in-time algorithm for approximating the solution of these two linear systems. Theuse of less parareal iterations makes it possible to reduce the time integration costs and to improve the time parallelscalability, and therefore, making it possible to really consider optimization in real time. We also show that truncatedmethods could be used without much delay in convergence, but with important savings in storage. Spectral boundsare provided and numerical experiments with the full orthogonalization method (FOM), and with inexact and trun-cated version of FOM are presented, illustrating the potential of the proposed methods.

Coauthor(s) : Xiuhong Du (du@alfred.edu) , Marcus Sarkis (msarkis@wpi.edu) , Christian E. Schaerer (cschaer@pol.una.py)

Speaker : Xiaocheng Tang (Lehigh University, xiaocheng.t@gmail.com)Title : Complexity of Inexact Proximal Newton methodAbstract : Recently several methods were proposed for sparse optimization which make careful use of second- order information

to improve local convergence rates. These methods construct a composite quadratic approximation using Hessianinformation, optimize this approximation using a first-order method, such as coordinate descent and employ a linesearch to ensure sufficient descent. Here we propose a general method, which improves upon these ideas to improvethe practical performance and prove a global convergence analysis in the spirit of proximal gradient methods.

Coauthor(s) : Katya Scheinberg (katyascheinberg@gmail.com)

Speaker : Aurelie Thiele (Lehigh University, aurelie.thiele@lehigh.edu)Title : Robust Partial CapitationAbstract : We investigate the optimal trade-off between fee-for-service and flat-fee payment (capitation) systems, the combi-

nation of which is recognized by health policy experts as having strong potential to slow rising healthcare costs. Weincorporate high cost uncertainty through a robust optimization approach and analyze several capitation models intheory and practice.

Speaker : Nghia Tran (Wayne State University, nghia@math.wayne.edu)Title : Full stability in nonlinear optimization with applications to semidefinite programmingAbstract : The paper concerns a systematic study of full stability in general optimization models including its conventional

Lipschitzian version as well as the new Hölderian one. We derive various characterizations of both Lipschitzian andHölderian full stability in nonsmooth optimization. The characterizations obtained are given in terms of second-order growth conditions and also via second-order generalized differential constructions of variational analysis. Wedevelop effective applications of our general characterizations of full stability to conventional models of nonlinearprogramming and semidefinite programming.

Coauthor(s) : Boris S. Mordukhovich (boris@math.wayne.edu)

Speaker : Robert Vanderbei (Princeton University, rvdb@princeton.edu)Title : Fast-Fourier OptimizationAbstract : Many interesting and fundamentally practical optimization problems involve constraints on the Fourier transform of

a function. It is well-known that the fast Fourier transform (fft) is a recursive algorithm that can dramatically improvethe efficiency for computing the discrete Fourier transform. In this paper, we explain the main idea behind the fastFourier transform and show how to adapt it in such a manner as to make it encodable as constraints in an optimizationproblem. We will show that the “fast Fourier” version of the optimization constraints produces a larger but sparserconstraint matrix and therefore one can think of the fast Fourier transform as a method of sparsifying the constraintsin an optimization problem. We demonstrate a real-world problem from the field of high-contrast imaging.

29

Speaker : Shuyi Wang (Lehigh University, shw210@lehigh.edu)Title : Robust Value-Based Insurance DesignAbstract : Value-based insurance design (VBID) encourages positive behavior from the insured by, for instance, providing pre-

ventive care for free or with lower coinsurance. We describe quantitative models to set cost-sharing levels for anEssential Benefit Package’s tiered benefits. We tie the probability of a patient undergoing a procedure to the finan-cial burden the coinsurance would represent, given income. We use robust optimization to capture uncertainty oncustomer behavior and procedure benefit.

Coauthor(s) : Aurélie Thiele (aut204@lehigh.edu)

Speaker : Jingjie Xiao (Purdue University, School of Industrial Engineering, jingjie.xiao@gmail.com)Title : Implementing Real-Time Pricing in Wholesale Electricity MarketsAbstract : A key hurdle for implementing real-time pricing of electricity is the lack of consumers’ responses. Solutions to over-

come the hurdle include energy management systems that can automatically optimize household appliance usagesuch as EV/PEV charge (and discharge with V2G) via two-way communication with the grid. One of the main goalsof this talk is to present how real-time retail pricing, aided by control automation devices, can be integrated intowholesale electricity markets under various uncertainties through approximate dynamic programming (ADP). Whatdistinguishes this paper from existing work in the literature is that wholesale electricity prices are endogenously de-termined as we have an explicit economic dispatch model embedded in the dynamic programming problem. ThisADP-based modeling framework will allow the feedback loop between electricity prices and electricity consumptionto be fully captured. We use deterministic linear programming benchmarks to demonstrate the quality of our ADPsolutions. The other goal of the talk is to use the modeling framework to provide numerical evidence to the debatethat if the dynamic rate structure is superior than the current flat rate structure in terms of both economic and envi-ronmental impacts.

Coauthor(s) : Andrew L. Liu (liu334@purdue.edu)

Speaker : Tengjiao Xiao (Lehigh University, tex209@lehigh.edu)Title : Robust Risk Adjustment in Health InsuranceAbstract : This paper introduces robust optimization models to address ambiguity and uncertainty in risk adjustment which is

used to quantify payment transfers across health plans. In particular, this paper develops robust scoring mechanisms,investigates the usefulness of having multiple risk scores instead of a single number, and tests the impact of the timehorizons involved. Learning is also incorporated in the risk adjustment models, since more information will becomeavailable over time for the newly insured when they submit claims.

Coauthor(s) : Aurelie Thiele (aut204@lehigh.edu)

Speaker : Sanjay Yadav (GL Noble Denton, sanjay.yadav@gl-group.com)Title : Piecewise-constant regression with implicit filteringAbstract : We will examine a piecewise-constant regression problem in the context of optimization of gas pipeline operations.

The decision variables in these optimization problems are pipeline operational parameters. When solving such prob-lems, it is often convenient to allow the decision variables to vary with arbitrary frequency. However, in practice,there are constraints on how frequently operational changes can be made in a pipeline. Once we have an optimalpipeline operational strategy, a more practical operational strategy can be constructed by fitting a piecewise-constantfunction to the original optimal strategy. In this piecewise-constant regression problem, the location and number ofbreakpoints are not known a priori. We will examine the use of implicit filtering optimization to solve this piecewise-constant regression problem.

Coauthor(s) : Richard Carter (richard.carter@gl-group.com) , Omar Hyjek (omar.hyjek@gl-group.com)

Speaker : Boshi Yang (The University of Iowa, boshi-yang@uiowa.edu)Title : The Trust Region Subproblem with Non-Intersecting Linear ConstraintsAbstract : This paper studies an extended trust region subproblem (eTRS) in which the trust region intersects the unit ball with

m linear inequality constraints. When m = 0, m = 1, or m = 2 and the linear constraints are parallel, it is known thatthe eTRS optimal value equals the optimal value of a particular convex relaxation, which is solvable in polynomialtime. However, it is also known that, when m ≥ 2 and at least two of the linear constraints intersect within the ball,i.e., some feasible point of the eTRS satisfies both linear constraints at equality, then the same convex relaxation mayadmit a gap with eTRS. This paper shows that the convex relaxation has no gap for arbitrary m as long as the linearconstraints are non-intersecting.

Coauthor(s) : Samuel Burer (samuel-burer@uiowa.edu)

30

Speaker : Yanhong Yang (Lehigh University, yay212@lehigh.edu)Title : Dynamic-programming-based Link Assignment for Data Collection in Wireless Sensor NetworksAbstract : As one of the basic applications of wireless sensor network (WSN), data collection has been discussed in the context

of miscellaneous application scenarios and topology configurations. When sensors are organized into a tree-basednetwork for data collection, the multi-channel characteristic of the transceivers can be utilized to reduce the proba-bility of collisions and improve system throughput. Joint Frequency Time Slot Scheduling (JFTSS) is considered thestate-of-the-art that approaches the optimal solution, but its number of slots grows rapidly when the link quality isunstable. In order to improve the link assignment schemes toward the optimal solution, we propose an algorithmthat further exploits the network topology and maximizes concurrent communication within each time slot throughdynamic programming (DP). Simulation results have shown that the number of required time slots increases slowerusing our algorithm when a more realistic model of link reliability is considered. In addition, the impacts of the loca-tion of the sink node and the communication range of the sensors are also analyzed.

Coauthor(s) : Liang Cheng (lica@lehigh.edu) , Huan Yang (huy213@lehigh.edu)

Speaker : Luis F. Zuluaga (Lehigh University, luis.zuluaga@lehigh.edu)Title : Extensions of Scarf’s max-min order formulaAbstract : Scarf’s max-min order formula is a classical result in the field of inventory management. It finds the order quantity

that minimizes the worst-case expected payoff of a single product inventory problem when only the mean and thevariance of the product’s demand distribution is assumed to be known. Specifically, we consider the case in which thedecision maker placing the order is, besides ambiguity-averse (i.e., assumes no complete knowledge of the product’sdemand distribution), also risk-averse; that is, besides the expected profit, he considers the risk associated with it interms of the profit’s standard deviation.

Coauthor(s) : Donglei Du (ddu@unb.ca) , Qiaoming Han (qmhan@nju.edu.cn)

31

32 !

!

ISE Centers

!

Enterprise Systems Center http://www.lehigh.edu/~inesc/ The Enterprise Systems Center is committed to helping students learn while simultaneously providing value for our clients. We believe that our research should be driven by industry needs and enabled by close partnerships and collaboration.

Center for Value Chain Research http://www.lehigh.edu/~inchain/ The Center for Value Chain Research (CVCR) is committed to promoting and conducting research and information exchange through the integration of emerging theory and best practices. The Center's research focuses primarily on value chain planning and development activities, which connect corporate strategy with value chain execution systems.

Computational Optimization Research at Lehigh http://coral.ie.lehigh.edu/ COR@L aims at promoting and conducting graduate-level research, primarily in the areas that lie at the interface of optimization and high-performance computing. Research conducted at the COR@L lab in recent years has focused on cutting edge optimization theory and development of several open source optimization software. The lab brings together faculty and graduate students aimed at establishing a multi-disciplinary research agenda. Research findings are disseminated through refereed publications, national and international conferences, and scholarly presentations.

33!!

Notes

____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

!

34!!

Notes

____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

!

Harold S. Mohler Laboratory

200 W. Packer Ave. Bethlehem, PA 18015

610-758-4050

Follow us! facebook.com/LehighISE

twitter.com/lehigh_ise

www.lehigh.edu/ise